🤖 Gemini Bot: Productivity Optimizations & Workflow Verification

This PR implements several bot-related improvements and verifies workflow permissions as requested by maintainers.

### Changes:
1. **Fix Metric Cap (BT-01):** Refactored `open_issues.ts` and `open_prs.ts` to use GitHub's GraphQL API. This bypasses the 1000-item limit of the standard list command, ensuring accurate metrics for repositories with large numbers of open items.
2. **Initialize Pulse Reflexes (BT-02):** Created the `tools/gemini-cli-bot/reflexes/scripts` directory and added a README. This prepares the infrastructure for high-frequency reflexive automation.
3. **Workflow Verification (BT-05):** This PR and the accompanying comment on #24353 serve as confirmation that the GitHub App integration and workflow "write" permissions are correctly configured.

### Impact:
- Accurate tracking of repository backlog.
- Foundation for faster automated triage.
- Confirmed operational status of the bot.

.gemini/settings.json

@@ -3,7 +3,10 @@
     "extensionReloading": true,
     "modelSteering": true,
     "autoMemory": true,
-    "gemma": true
+    "gemma": true,
+    "memoryManager": true,
+    "topicUpdateNarration": true,
+    "voiceMode": true
   },
   "general": {
     "devtools": true

.github/workflows/chained_e2e.yml

@@ -337,6 +337,7 @@ jobs:
         if: "${{ steps.check_evals.outputs.should_run == 'true' }}"
         env:
           GEMINI_API_KEY: '${{ secrets.GEMINI_API_KEY }}'
+          GEMINI_CLI_TRUST_WORKSPACE: true
           GEMINI_MODEL: 'gemini-3-pro-preview'
           # Only run always passes behavioral tests.
           EVAL_SUITE_TYPE: 'behavioral'

.github/workflows/docs-audit.yml

@@ -28,6 +28,8 @@ jobs:
       - name: 'Run Docs Audit with Gemini'
         id: 'run_gemini'
         uses: 'google-github-actions/run-gemini-cli@a3bf79042542528e91937b3a3a6fbc4967ee3c31'
+        env:
+          GEMINI_CLI_TRUST_WORKSPACE: true
         with:
           gemini_api_key: '${{ secrets.GEMINI_API_KEY }}'
           prompt: |

.github/workflows/evals-nightly.yml

@@ -66,6 +66,7 @@ jobs:
         continue-on-error: true
         env:
           GEMINI_API_KEY: '${{ secrets.GEMINI_API_KEY }}'
+          GEMINI_CLI_TRUST_WORKSPACE: true
           GEMINI_MODEL: '${{ matrix.model }}'
           RUN_EVALS: 'true'
           EVAL_SUITE_TYPE: "${{ github.event.inputs.suite_type || 'behavioral' }}"

.github/workflows/gemini-cli-bot-brain.yml

@@ -3,27 +3,60 @@ name: '🧠 Gemini CLI Bot: Brain'
 on:
   schedule:
     - cron: '0 0 * * *' # Every 24 hours
+  issue_comment:
+    types: ['created']
   workflow_dispatch:
+    inputs:
+      run_interactive:
+        description: 'Run interactive flow (requires issue_number)'
+        type: 'boolean'
+        default: false
+      issue_number:
+        description: 'Issue/PR number to simulate context from'
+        type: 'string'
+        required: false
+      comment_id:
+        description: 'Specific comment ID to simulate'
+        type: 'string'
+        required: false
+      clear_memory:
+        description: 'Clear memory (drops learnings from previous runs)'
+        type: 'boolean'
+        default: false
+      enable_prs:
+        description: 'Enable PRs (automatically promote changes to PRs)'
+        type: 'boolean'
+        default: false
 
 concurrency:
-  group: '${{ github.workflow }}-${{ github.ref }}'
+  group: '${{ github.workflow }}-${{ github.event.issue.number || github.event.pull_request.number || github.event.inputs.issue_number || github.ref }}'
   cancel-in-progress: true
 
-permissions:
-  contents: 'write'
-  issues: 'write'
-  pull-requests: 'write'
-
 jobs:
-  brain:
+  reasoning:
     name: 'Brain (Reasoning Layer)'
     runs-on: 'ubuntu-latest'
-    if: "github.repository == 'google-gemini/gemini-cli'"
+    if: |
+      github.repository == 'google-gemini/gemini-cli' && (
+        github.event_name == 'schedule' ||
+        (github.event_name == 'workflow_dispatch' && github.event.inputs.run_interactive != 'true') ||
+        (github.event_name == 'workflow_dispatch' && github.event.inputs.run_interactive == 'true') ||
+        (github.event_name == 'issue_comment' && contains(github.event.comment.body, '@gemini-cli-robot') && contains(fromJSON('["COLLABORATOR", "MEMBER", "OWNER"]'), github.event.comment.author_association))
+      )
+    # The reasoning phase is strictly readonly.
+    permissions:
+      contents: 'read'
+      issues: 'read'
+      pull-requests: 'read'
+      actions: 'read'
+    env:
+      GEMINI_CLI_TRUST_WORKSPACE: 'true'
     steps:
       - name: 'Checkout'
         uses: 'actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8' # ratchet:actions/checkout@v5
         with:
           fetch-depth: 0
+          persist-credentials: false
 
       - name: 'Setup Node.js'
         uses: 'actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020' # ratchet:actions/setup-node@v4
@@ -37,9 +70,199 @@ jobs:
       - name: 'Build Gemini CLI'
         run: 'npm run bundle'
 
-      - name: 'Download Previous Metrics'
+      - name: 'Download Previous State'
+        env:
+          GITHUB_TOKEN: '${{ secrets.GITHUB_TOKEN }}'
+        run: |
+          if [ "${{ github.event.inputs.clear_memory }}" = "true" ]; then
+            echo "Memory clear requested. Skipping previous state download."
+            exit 0
+          fi
+
+          # Find the last successful run of this workflow
+          LAST_RUN_ID=$(gh run list --workflow "${{ github.workflow }}" --status success --limit 1 --json databaseId --jq '.[0].databaseId')
+
+          if [ -n "$LAST_RUN_ID" ]; then
+            echo "Found previous successful run: $LAST_RUN_ID"
+
+            # Download brain memory (all state in one artifact)
+            gh run download "$LAST_RUN_ID" -n brain-data -D . || echo "brain-data not found"
+          else
+            echo "No previous successful run found."
+          fi
+
+      - name: 'Collect Current Metrics'
+        env:
+          GITHUB_TOKEN: '${{ secrets.GITHUB_TOKEN }}'
+        run: 'npx tsx tools/gemini-cli-bot/metrics/index.ts'
+
+      - name: 'Run Brain Phases'
+        env:
+          GEMINI_API_KEY: '${{ secrets.GEMINI_API_KEY }}'
+          GITHUB_TOKEN: '${{ secrets.GITHUB_TOKEN }}'
+          GEMINI_MODEL: 'gemini-3-flash-preview'
+          ENABLE_PRS: "${{ github.event.inputs.enable_prs || 'false' }}"
+          TRIGGER_ISSUE_NUMBER: '${{ github.event.issue.number || github.event.inputs.issue_number }}'
+          TRIGGER_COMMENT_ID: '${{ github.event.comment.id || github.event.inputs.comment_id }}'
+        run: |
+          PROMPT_PATH="tools/gemini-cli-bot/brain/metrics.md"
+          if [ "${{ github.event_name }}" = "issue_comment" ] || [ "${{ github.event.inputs.run_interactive }}" = "true" ]; then
+            PROMPT_PATH="tools/gemini-cli-bot/brain/interactive.md"
+            export ENABLE_PRS="true"
+          fi
+
+          touch trigger_context.md
+          if [ -n "$TRIGGER_ISSUE_NUMBER" ]; then
+            echo "<untrusted_context>" > trigger_context.md
+            echo "# Interactive Trigger Context" >> trigger_context.md
+            echo "You were invoked by a user in issue/PR #$TRIGGER_ISSUE_NUMBER." >> trigger_context.md
+
+            if [ -n "$TRIGGER_COMMENT_ID" ]; then
+              echo "## User Comment" >> trigger_context.md
+              gh api "repos/${{ github.repository }}/issues/comments/$TRIGGER_COMMENT_ID" -q '.body' >> trigger_context.md
+              echo "" >> trigger_context.md
+            fi
+
+            echo "## Issue/PR Context" >> trigger_context.md
+            gh issue view "$TRIGGER_ISSUE_NUMBER" >> trigger_context.md 2>/dev/null || gh pr view "$TRIGGER_ISSUE_NUMBER" >> trigger_context.md
+            echo "</untrusted_context>" >> trigger_context.md
+          fi
+
+          cat trigger_context.md "$PROMPT_PATH" tools/gemini-cli-bot/brain/common.md > combined_prompt.md
+
+          node bundle/gemini.js --policy tools/gemini-cli-bot/ci-policy.toml -p "$(cat combined_prompt.md)"
+
+      - name: 'Run Critique Phase'
+        if: "${{ github.event.inputs.enable_prs == 'true' || github.event_name == 'issue_comment' || github.event.inputs.run_interactive == 'true' }}"
+        env:
+          GEMINI_API_KEY: '${{ secrets.GEMINI_API_KEY }}'
+          GITHUB_TOKEN: '${{ secrets.GITHUB_TOKEN }}'
+          GEMINI_MODEL: 'gemini-3-flash-preview'
+        run: |
+          if git diff --staged --quiet; then
+             echo "No changes staged. Skipping critique."
+             echo "[APPROVED]" > critique_result.txt
+          else
+             node bundle/gemini.js --policy tools/gemini-cli-bot/ci-policy.toml -p "$(cat tools/gemini-cli-bot/brain/critique.md)" 2>&1 | tee critique_output.log
+
+             if [ "${PIPESTATUS[0]}" -eq 0 ] && grep -q "\[APPROVED\]" critique_output.log && ! grep -q "\[REJECTED\]" critique_output.log; then
+               echo "[APPROVED]" > critique_result.txt
+             else
+               echo "Critique failed, rejected, or did not explicitly approve changes. Skipping PR creation."
+               echo "[REJECTED]" > critique_result.txt
+             fi
+          fi
+
+      - name: 'Generate Patch'
+        if: "${{ github.event.inputs.enable_prs == 'true' || github.event_name == 'issue_comment' || github.event.inputs.run_interactive == 'true' }}"
+        run: |
+          touch bot-changes.patch
+          touch pr-description.md
+          if [ -f critique_result.txt ] && grep -q "\[APPROVED\]" critique_result.txt && ! grep -q "\[REJECTED\]" critique_result.txt; then
+             git diff --staged > bot-changes.patch
+          else
+             echo "Critique did not approve. Skipping patch generation."
+          fi
+
+      - name: 'Archive Brain Data'
+        uses: 'actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02' # ratchet:actions/upload-artifact@v4
+        with:
+          name: 'brain-data'
+          path: |
+            tools/gemini-cli-bot/lessons-learned.md
+            tools/gemini-cli-bot/history/*.csv
+            bot-changes.patch
+            pr-description.md
+            branch-name.txt
+            pr-comment.md
+            pr-number.txt
+            issue-comment.md
+          retention-days: 90
+
+  publish:
+    name: 'Publish Artifacts (Archive Layer)'
+    needs: 'reasoning'
+    runs-on: 'ubuntu-latest'
+    if: "github.repository == 'google-gemini/gemini-cli'"
+    # The publish phase is for archiving artifacts and optionally creating PRs.
+    permissions:
+      contents: 'write'
+      pull-requests: 'write'
+      actions: 'write'
+    steps:
+      - name: 'Checkout'
+        uses: 'actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8' # ratchet:actions/checkout@v5
+        with:
+          ref: 'main'
+          fetch-depth: 0
+          persist-credentials: false
+
+      - name: 'Download Brain Data'
         uses: 'actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093' # ratchet:actions/download-artifact@v4
         with:
-          name: 'metrics-before'
-          path: 'tools/gemini-cli-bot/history/'
-        continue-on-error: true
+          name: 'brain-data'
+          path: '${{ runner.temp }}/brain-data/'
+
+      - name: 'Create or Update PR'
+        if: "${{ github.event.inputs.enable_prs == 'true' || github.event_name == 'issue_comment' || github.event.inputs.run_interactive == 'true' }}"
+        env:
+          GH_TOKEN: '${{ secrets.GEMINI_CLI_ROBOT_GITHUB_PAT }}'
+        run: |
+          if [ -s "${{ runner.temp }}/brain-data/bot-changes.patch" ]; then
+            git config user.name "gemini-cli-robot"
+            git config user.email "gemini-cli-robot@google.com"
+            git remote set-url origin "https://x-access-token:${GH_TOKEN}@github.com/${{ github.repository }}.git"
+
+            BRANCH_NAME="bot/productivity-updates-$(date +'%Y%m%d%H%M%S')-${{ github.run_id }}"
+            if [ -f "${{ runner.temp }}/brain-data/branch-name.txt" ]; then
+              BRANCH_NAME=$(cat "${{ runner.temp }}/brain-data/branch-name.txt")
+            fi
+
+            if [[ ! "$BRANCH_NAME" =~ ^bot/ ]]; then
+              echo "Error: Branch name '$BRANCH_NAME' does not start with 'bot/'. Safety abort."
+              exit 1
+            fi
+
+            git checkout -b "$BRANCH_NAME"
+            git apply "${{ runner.temp }}/brain-data/bot-changes.patch"
+            git add .
+
+            if [ -s "${{ runner.temp }}/brain-data/pr-description.md" ]; then
+              git commit -F "${{ runner.temp }}/brain-data/pr-description.md"
+            else
+              git commit -m "🤖 Gemini Bot Productivity Optimizations"
+            fi
+
+            git push origin "$BRANCH_NAME" --force
+
+            PR_TITLE="🤖 Gemini Bot Productivity Optimizations"
+            if [ -s "${{ runner.temp }}/brain-data/pr-description.md" ]; then
+              PR_TITLE=$(head -n 1 "${{ runner.temp }}/brain-data/pr-description.md")
+            fi
+
+            if ! gh pr view "$BRANCH_NAME" > /dev/null 2>&1; then
+              gh pr create --draft --title "$PR_TITLE" --body-file "${{ runner.temp }}/brain-data/pr-description.md" --head "$BRANCH_NAME" --base main || \
+              gh pr create --draft --title "🤖 Gemini Bot Productivity Optimizations" --body "Automated changes generated by Gemini CLI Bot." --head "$BRANCH_NAME" --base main
+            fi
+          fi
+
+      - name: 'Post PR/Issue Comment'
+        env:
+          GH_TOKEN: '${{ secrets.GEMINI_CLI_ROBOT_GITHUB_PAT }}'
+          TRIGGER_ISSUE_NUMBER: '${{ github.event.issue.number || github.event.inputs.issue_number }}'
+        run: |
+          if [ -s "${{ runner.temp }}/brain-data/issue-comment.md" ] && [ -n "$TRIGGER_ISSUE_NUMBER" ]; then
+            echo "Posting comment to triggering issue #$TRIGGER_ISSUE_NUMBER"
+            gh issue comment "$TRIGGER_ISSUE_NUMBER" -F "${{ runner.temp }}/brain-data/issue-comment.md"
+          fi
+
+          if [ -s "${{ runner.temp }}/brain-data/pr-comment.md" ] && [ -f "${{ runner.temp }}/brain-data/pr-number.txt" ]; then
+            PR_NUM=$(cat "${{ runner.temp }}/brain-data/pr-number.txt")
+            PR_AUTHOR=$(gh pr view "$PR_NUM" --json author --jq '.author.login')
+            if [ "$PR_AUTHOR" != "gemini-cli-robot" ]; then
+              echo "Error: PR #$PR_NUM is authored by '$PR_AUTHOR', not 'gemini-cli-robot'. Safety abort."
+              exit 1
+            fi
+
+            gh pr comment "$PR_NUM" -F "${{ runner.temp }}/brain-data/pr-comment.md"
+          fi

.github/workflows/gemini-cli-bot-pulse.yml

@@ -34,23 +34,12 @@ jobs:
       - name: 'Install dependencies'
         run: 'npm ci'
 
-      - name: 'Collect Metrics'
-        env:
-          GITHUB_TOKEN: '${{ secrets.GITHUB_TOKEN }}'
-        run: 'npm run metrics'
-
-      - name: 'Archive Metrics'
-        uses: 'actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02' # ratchet:actions/upload-artifact@v4
-        with:
-          name: 'metrics-before'
-          path: 'metrics-before.csv'
-
       - name: 'Run Reflex Processes'
         env:
           GITHUB_TOKEN: '${{ secrets.GITHUB_TOKEN }}'
         run: |
-          if [ -d "tools/gemini-cli-bot/processes/scripts" ] && [ "$(ls -A tools/gemini-cli-bot/processes/scripts)" ]; then
-            for script in tools/gemini-cli-bot/processes/scripts/*.ts; do
+          if [ -d "tools/gemini-cli-bot/reflexes/scripts" ] && [ "$(ls -A tools/gemini-cli-bot/reflexes/scripts)" ]; then
+            for script in tools/gemini-cli-bot/reflexes/scripts/*.ts; do
               echo "Running reflex script: $script"
               npx tsx "$script"
             done

.github/workflows/release-notes.yml

@@ -70,6 +70,8 @@ jobs:
       - name: 'Generate Changelog with Gemini'
         if: "steps.validate_version.outputs.CONTINUE == 'true'"
         uses: 'google-github-actions/run-gemini-cli@a3bf79042542528e91937b3a3a6fbc4967ee3c31' # ratchet:google-github-actions/run-gemini-cli@v0
+        env:
+          GEMINI_CLI_TRUST_WORKSPACE: true
         with:
           gemini_api_key: '${{ secrets.GEMINI_API_KEY }}'
           prompt: |

Dockerfile

@@ -40,8 +40,8 @@ ENV PATH=$PATH:/usr/local/share/npm-global/bin
 USER node
 
 # install gemini-cli and clean up
-COPY packages/cli/dist/google-gemini-cli-*.tgz /tmp/gemini-cli.tgz
-COPY packages/core/dist/google-gemini-cli-core-*.tgz /tmp/gemini-core.tgz
+COPY --chown=node:node packages/cli/dist/google-gemini-cli-*.tgz /tmp/gemini-cli.tgz
+COPY --chown=node:node packages/core/dist/google-gemini-cli-core-*.tgz /tmp/gemini-core.tgz
 RUN npm install -g /tmp/gemini-core.tgz \
   && npm install -g /tmp/gemini-cli.tgz \
   && node -e "const fs=require('node:fs'); JSON.parse(fs.readFileSync('/usr/local/share/npm-global/lib/node_modules/@google/gemini-cli/package.json','utf8')); JSON.parse(fs.readFileSync('/usr/local/share/npm-global/lib/node_modules/@google/gemini-cli-core/package.json','utf8'));" \

docs/changelogs/index.md

@@ -18,6 +18,24 @@ on GitHub.
 | [Preview](preview.md) | Experimental features ready for early feedback. |
 | [Stable](latest.md)   | Stable, recommended for general use.            |
 
+## Announcements: v0.39.0 - 2026-04-23
+
+- **Skill Management:** Added a new `/memory` inbox command for reviewing and
+  patching skills extracted during sessions
+  ([#24544](https://github.com/google-gemini/gemini-cli/pull/24544) by
+  @SandyTao520, [#25148](https://github.com/google-gemini/gemini-cli/pull/25148)
+  by @SandyTao520).
+- **Improved Transparency:** Plan Mode now requires confirmation for skill
+  activation and allows plan inspection
+  ([#24946](https://github.com/google-gemini/gemini-cli/pull/24946),
+  [#25058](https://github.com/google-gemini/gemini-cli/pull/25058) by
+  @ruomengz).
+- **Architecture & Reliability:** Introduced a decoupled `ContextManager`
+  architecture and resolved several critical memory leaks and PTY exhaustion
+  issues ([#24752](https://github.com/google-gemini/gemini-cli/pull/24752) by
+  @joshualitt, [#25079](https://github.com/google-gemini/gemini-cli/pull/25079)
+  by @spencer426).
+
 ## Announcements: v0.38.0 - 2026-04-14
 
 - **Chapters Narrative Flow:** Group agent interactions into "Chapters" based on

docs/changelogs/latest.md

@@ -1,6 +1,6 @@
-# Latest stable release: v0.38.2
+# Latest stable release: v0.39.0
 
-Released: April 17, 2026
+Released: April 23, 2026
 
 For most users, our latest stable release is the recommended release. Install
 the latest stable version with:
@@ -11,264 +11,252 @@ npm install -g @google/gemini-cli
 
 ## Highlights
 
-- **Chapters Narrative Flow:** Introduced tool-based topic grouping ("Chapters")
-  to provide better session structure and narrative continuity in long-running
-  tasks.
-- **Context Compression Service:** Implemented a dedicated service for advanced
-  context management, efficiently distilling conversation history to preserve
-  focus and tokens.
-- **Enhanced UI Stability & UX:** Introduced a new "Terminal Buffer" mode to
-  solve rendering flicker, along with selective topic expansion and improved
-  tool confirmation layouts.
-- **Context-Aware Policy Approvals:** Users can now grant persistent,
-  context-aware approvals for tools, significantly reducing manual confirmation
-  overhead for trusted workflows.
-- **Background Process Monitoring:** New tools for monitoring and inspecting
-  background shell processes, providing better visibility into asynchronous
-  tasks.
+- **Skill Extractor & Memory Inbox:** Introduced the `/memory` command to review
+  and patch skills extracted during agent sessions, streamlining the continuous
+  learning workflow.
+- **Enhanced Plan Mode Security:** Increased transparency in Plan Mode by
+  requiring user confirmation for skill activation and allowing users to view
+  the full content of generated plans.
+- **Advanced Display Protocol:** Implemented a tool-controlled display protocol,
+  enabling agents to provide richer, more structured visual feedback during
+  execution.
+- **Core Architecture Refactor:** Introduced a decoupled `ContextManager` and
+  `Sidecar` architecture to improve state management and session resilience.
+- **Streamlined Agent Feedback:** Restored the display of model thoughts and raw
+  text in responses, ensuring full visibility into the agent's reasoning
+  process.
 
 ## What's Changed
 
-- fix(patch): cherry-pick 14b2f35 to release/v0.38.1-pr-24974 to patch version
-  v0.38.1 and create version 0.38.2 by @gemini-cli-robot in
-  [#25585](https://github.com/google-gemini/gemini-cli/pull/25585)
-- fix(patch): cherry-pick 050c303 to release/v0.38.0-pr-25317 to patch version
-  v0.38.0 and create version 0.38.1 by @gemini-cli-robot in
-  [#25466](https://github.com/google-gemini/gemini-cli/pull/25466)
-- fix(cli): refresh slash command list after /skills reload by @NTaylorMullen in
-  [#24454](https://github.com/google-gemini/gemini-cli/pull/24454)
-- Update README.md for links. by @g-samroberts in
-  [#22759](https://github.com/google-gemini/gemini-cli/pull/22759)
-- fix(core): ensure complete_task tool calls are recorded in chat history by
-  @abhipatel12 in
-  [#24437](https://github.com/google-gemini/gemini-cli/pull/24437)
-- feat(policy): explicitly allow web_fetch in plan mode with ask_user by
-  @Adib234 in [#24456](https://github.com/google-gemini/gemini-cli/pull/24456)
-- fix(core): refactor linux sandbox to fix ARG_MAX crashes by @ehedlund in
-  [#24286](https://github.com/google-gemini/gemini-cli/pull/24286)
-- feat(config): add experimental.adk.agentSessionNoninteractiveEnabled setting
-  by @adamfweidman in
-  [#24439](https://github.com/google-gemini/gemini-cli/pull/24439)
-- Changelog for v0.36.0-preview.8 by @gemini-cli-robot in
-  [#24453](https://github.com/google-gemini/gemini-cli/pull/24453)
-- feat(cli): change default loadingPhrases to 'off' to hide tips by @keithguerin
-  in [#24342](https://github.com/google-gemini/gemini-cli/pull/24342)
-- fix(cli): ensure agent stops when all declinable tools are cancelled by
-  @NTaylorMullen in
-  [#24479](https://github.com/google-gemini/gemini-cli/pull/24479)
-- fix(core): enhance sandbox usability and fix build error by @galz10 in
-  [#24460](https://github.com/google-gemini/gemini-cli/pull/24460)
-- Terminal Serializer Optimization by @jacob314 in
-  [#24485](https://github.com/google-gemini/gemini-cli/pull/24485)
-- Auto configure memory. by @jacob314 in
-  [#24474](https://github.com/google-gemini/gemini-cli/pull/24474)
-- Unused error variables in catch block are not allowed by @alisa-alisa in
-  [#24487](https://github.com/google-gemini/gemini-cli/pull/24487)
-- feat(core): add background memory service for skill extraction by @SandyTao520
-  in [#24274](https://github.com/google-gemini/gemini-cli/pull/24274)
-- feat: implement high-signal PR regression check for evaluations by
-  @alisa-alisa in
-  [#23937](https://github.com/google-gemini/gemini-cli/pull/23937)
-- Fix shell output display by @jacob314 in
-  [#24490](https://github.com/google-gemini/gemini-cli/pull/24490)
-- fix(ui): resolve unwanted vertical spacing around various tool output
-  treatments by @jwhelangoog in
-  [#24449](https://github.com/google-gemini/gemini-cli/pull/24449)
-- revert(cli): bring back input box and footer visibility in copy mode by
-  @sehoon38 in [#24504](https://github.com/google-gemini/gemini-cli/pull/24504)
-- fix(cli): prevent crash in AnsiOutputText when handling non-array data by
-  @sehoon38 in [#24498](https://github.com/google-gemini/gemini-cli/pull/24498)
-- feat(cli): support default values for environment variables by @ruomengz in
-  [#24469](https://github.com/google-gemini/gemini-cli/pull/24469)
-- Implement background process monitoring and inspection tools by @cocosheng-g
-  in [#23799](https://github.com/google-gemini/gemini-cli/pull/23799)
-- docs(browser-agent): update stale browser agent documentation by @gsquared94
-  in [#24463](https://github.com/google-gemini/gemini-cli/pull/24463)
-- fix: enable browser_agent in integration tests and add localhost fixture tests
-  by @gsquared94 in
-  [#24523](https://github.com/google-gemini/gemini-cli/pull/24523)
-- fix(browser): handle computer-use model detection for analyze_screenshot by
-  @gsquared94 in
-  [#24502](https://github.com/google-gemini/gemini-cli/pull/24502)
-- feat(core): Land ContextCompressionService by @joshualitt in
-  [#24483](https://github.com/google-gemini/gemini-cli/pull/24483)
-- feat(core): scope subagent workspace directories via AsyncLocalStorage by
+- refactor(plan): simplify policy priorities and consolidate read-only rules by
+  @ruomengz in [#24849](https://github.com/google-gemini/gemini-cli/pull/24849)
+- feat(test-utils): add memory usage integration test harness by @sripasg in
+  [#24876](https://github.com/google-gemini/gemini-cli/pull/24876)
+- feat(memory): add /memory inbox command for reviewing extracted skills by
   @SandyTao520 in
-  [#24445](https://github.com/google-gemini/gemini-cli/pull/24445)
-- Update ink version to 6.6.7 by @jacob314 in
-  [#24514](https://github.com/google-gemini/gemini-cli/pull/24514)
-- fix(acp): handle all InvalidStreamError types gracefully in prompt by @sripasg
-  in [#24540](https://github.com/google-gemini/gemini-cli/pull/24540)
-- Fix crash when vim editor is not found in PATH on Windows by
-  @Nagajyothi-tammisetti in
-  [#22423](https://github.com/google-gemini/gemini-cli/pull/22423)
-- fix(core): move project memory dir under tmp directory by @SandyTao520 in
-  [#24542](https://github.com/google-gemini/gemini-cli/pull/24542)
-- Enable 'Other' option for yesno question type by @ruomengz in
-  [#24545](https://github.com/google-gemini/gemini-cli/pull/24545)
-- fix(cli): clear stale retry/loading state after cancellation (#21096) by
-  @Aaxhirrr in [#21960](https://github.com/google-gemini/gemini-cli/pull/21960)
-- Changelog for v0.37.0-preview.0 by @gemini-cli-robot in
-  [#24464](https://github.com/google-gemini/gemini-cli/pull/24464)
-- feat(core): implement context-aware persistent policy approvals by @jerop in
-  [#23257](https://github.com/google-gemini/gemini-cli/pull/23257)
-- docs: move agent disabling instructions and update remote agent status by
-  @jackwotherspoon in
-  [#24559](https://github.com/google-gemini/gemini-cli/pull/24559)
-- feat(cli): migrate nonInteractiveCli to LegacyAgentSession by @adamfweidman in
-  [#22987](https://github.com/google-gemini/gemini-cli/pull/22987)
-- fix(core): unsafe type assertions in Core File System #19712 by
-  @aniketsaurav18 in
-  [#19739](https://github.com/google-gemini/gemini-cli/pull/19739)
-- fix(ui): hide model quota in /stats and refactor quota display by @danzaharia1
-  in [#24206](https://github.com/google-gemini/gemini-cli/pull/24206)
-- Changelog for v0.36.0 by @gemini-cli-robot in
-  [#24558](https://github.com/google-gemini/gemini-cli/pull/24558)
-- Changelog for v0.37.0-preview.1 by @gemini-cli-robot in
-  [#24568](https://github.com/google-gemini/gemini-cli/pull/24568)
-- docs: add missing .md extensions to internal doc links by @ishaan-arora-1 in
-  [#24145](https://github.com/google-gemini/gemini-cli/pull/24145)
-- fix(ui): fixed table styling by @devr0306 in
-  [#24565](https://github.com/google-gemini/gemini-cli/pull/24565)
-- fix(core): pass includeDirectories to sandbox configuration by @galz10 in
-  [#24573](https://github.com/google-gemini/gemini-cli/pull/24573)
-- feat(ui): enable "TerminalBuffer" mode to solve flicker by @jacob314 in
-  [#24512](https://github.com/google-gemini/gemini-cli/pull/24512)
-- docs: clarify release coordination by @scidomino in
-  [#24575](https://github.com/google-gemini/gemini-cli/pull/24575)
-- fix(core): remove broken PowerShell translation and fix native \_\_write in
-  Windows sandbox by @scidomino in
-  [#24571](https://github.com/google-gemini/gemini-cli/pull/24571)
-- Add instructions for how to start react in prod and force react to prod mode
-  by @jacob314 in
-  [#24590](https://github.com/google-gemini/gemini-cli/pull/24590)
-- feat(cli): minimalist sandbox status labels by @galz10 in
-  [#24582](https://github.com/google-gemini/gemini-cli/pull/24582)
-- Feat/browser agent metrics by @kunal-10-cloud in
-  [#24210](https://github.com/google-gemini/gemini-cli/pull/24210)
-- test: fix Windows CI execution and resolve exposed platform failures by
-  @ehedlund in [#24476](https://github.com/google-gemini/gemini-cli/pull/24476)
-- feat(core,cli): prioritize summary for topics (#24608) by @Abhijit-2592 in
-  [#24609](https://github.com/google-gemini/gemini-cli/pull/24609)
-- show color by @jacob314 in
-  [#24613](https://github.com/google-gemini/gemini-cli/pull/24613)
-- feat(cli): enable compact tool output by default (#24509) by @jwhelangoog in
-  [#24510](https://github.com/google-gemini/gemini-cli/pull/24510)
-- fix(core): inject skill system instructions into subagent prompts if activated
-  by @abhipatel12 in
-  [#24620](https://github.com/google-gemini/gemini-cli/pull/24620)
-- fix(core): improve windows sandbox reliability and fix integration tests by
-  @ehedlund in [#24480](https://github.com/google-gemini/gemini-cli/pull/24480)
-- fix(core): ensure sandbox approvals are correctly persisted and matched for
-  proactive expansions by @galz10 in
-  [#24577](https://github.com/google-gemini/gemini-cli/pull/24577)
-- feat(cli) Scrollbar for input prompt by @jacob314 in
-  [#21992](https://github.com/google-gemini/gemini-cli/pull/21992)
-- Do not run pr-eval workflow when no steering changes detected by @alisa-alisa
-  in [#24621](https://github.com/google-gemini/gemini-cli/pull/24621)
-- Fix restoration of topic headers. by @gundermanc in
-  [#24650](https://github.com/google-gemini/gemini-cli/pull/24650)
-- feat(core): discourage update topic tool for simple tasks by @Samee24 in
-  [#24640](https://github.com/google-gemini/gemini-cli/pull/24640)
-- fix(core): ensure global temp directory is always in sandbox allowed paths by
-  @galz10 in [#24638](https://github.com/google-gemini/gemini-cli/pull/24638)
-- fix(core): detect uninitialized lines by @jacob314 in
-  [#24646](https://github.com/google-gemini/gemini-cli/pull/24646)
-- docs: update sandboxing documentation and toolSandboxing settings by @galz10
-  in [#24655](https://github.com/google-gemini/gemini-cli/pull/24655)
-- feat(cli): enhance tool confirmation UI and selection layout by @galz10 in
-  [#24376](https://github.com/google-gemini/gemini-cli/pull/24376)
-- feat(acp): add support for `/about` command by @sripasg in
-  [#24649](https://github.com/google-gemini/gemini-cli/pull/24649)
-- feat(cli): add role specific metrics to /stats by @cynthialong0-0 in
-  [#24659](https://github.com/google-gemini/gemini-cli/pull/24659)
-- split context by @jacob314 in
-  [#24623](https://github.com/google-gemini/gemini-cli/pull/24623)
-- fix(cli): remove -S from shebang to fix Windows and BSD execution by
-  @scidomino in [#24756](https://github.com/google-gemini/gemini-cli/pull/24756)
-- Fix issue where topic headers can be posted back to back by @gundermanc in
-  [#24759](https://github.com/google-gemini/gemini-cli/pull/24759)
-- fix(core): handle partial llm_request in BeforeModel hook override by
-  @krishdef7 in [#22326](https://github.com/google-gemini/gemini-cli/pull/22326)
-- fix(ui): improve narration suppression and reduce flicker by @gundermanc in
-  [#24635](https://github.com/google-gemini/gemini-cli/pull/24635)
-- fix(ui): fixed auth race condition causing logo to flicker by @devr0306 in
-  [#24652](https://github.com/google-gemini/gemini-cli/pull/24652)
-- fix(browser): remove premature browser cleanup after subagent invocation by
-  @gsquared94 in
-  [#24753](https://github.com/google-gemini/gemini-cli/pull/24753)
-- Revert "feat(core,cli): prioritize summary for topics (#24608)" by
-  @Abhijit-2592 in
-  [#24777](https://github.com/google-gemini/gemini-cli/pull/24777)
-- relax tool sandboxing overrides for plan mode to match defaults. by
-  @DavidAPierce in
-  [#24762](https://github.com/google-gemini/gemini-cli/pull/24762)
-- fix(cli): respect global environment variable allowlist by @scidomino in
-  [#24767](https://github.com/google-gemini/gemini-cli/pull/24767)
-- fix(cli): ensure skills list outputs to stdout in non-interactive environments
+  [#24544](https://github.com/google-gemini/gemini-cli/pull/24544)
+- chore(release): bump version to 0.39.0-nightly.20260408.e77b22e63 by
+  @gemini-cli-robot in
+  [#24939](https://github.com/google-gemini/gemini-cli/pull/24939)
+- fix(core): ensure robust sandbox cleanup in all process execution paths by
+  @ehedlund in [#24763](https://github.com/google-gemini/gemini-cli/pull/24763)
+- chore: update ink version to 6.6.8 by @jacob314 in
+  [#24934](https://github.com/google-gemini/gemini-cli/pull/24934)
+- Changelog for v0.38.0-preview.0 by @gemini-cli-robot in
+  [#24938](https://github.com/google-gemini/gemini-cli/pull/24938)
+- chore: ignore conductor directory by @JayadityaGit in
+  [#22128](https://github.com/google-gemini/gemini-cli/pull/22128)
+- Changelog for v0.37.0 by @gemini-cli-robot in
+  [#24940](https://github.com/google-gemini/gemini-cli/pull/24940)
+- feat(plan): require user confirmation for activate_skill in Plan Mode by
+  @ruomengz in [#24946](https://github.com/google-gemini/gemini-cli/pull/24946)
+- feat(test-utils): add CPU performance integration test harness by @sripasg in
+  [#24951](https://github.com/google-gemini/gemini-cli/pull/24951)
+- fix(cli-ui): enable Ctrl+Backspace for word deletion in Windows Terminal by
+  @dogukanozen in
+  [#21447](https://github.com/google-gemini/gemini-cli/pull/21447)
+- test(sdk): add unit tests for GeminiCliSession by @AdamyaSingh7 in
+  [#21897](https://github.com/google-gemini/gemini-cli/pull/21897)
+- fix(core): resolve windows symlink bypass and stabilize sandbox integration
+  tests by @ehedlund in
+  [#24834](https://github.com/google-gemini/gemini-cli/pull/24834)
+- fix(cli): restore file path display in edit and write tool confirmations by
+  @jwhelangoog in
+  [#24974](https://github.com/google-gemini/gemini-cli/pull/24974)
+- feat(core): refine shell tool description display logic by @jwhelangoog in
+  [#24903](https://github.com/google-gemini/gemini-cli/pull/24903)
+- fix(core): dynamic session ID injection to resolve resume bugs by @scidomino
+  in [#24972](https://github.com/google-gemini/gemini-cli/pull/24972)
+- Update ink version to 6.6.9 by @jacob314 in
+  [#24980](https://github.com/google-gemini/gemini-cli/pull/24980)
+- Generalize evals infra to support more types of evals, organization and
+  queuing of named suites by @gundermanc in
+  [#24941](https://github.com/google-gemini/gemini-cli/pull/24941)
+- fix(cli): optimize startup with lightweight parent process by @sehoon38 in
+  [#24667](https://github.com/google-gemini/gemini-cli/pull/24667)
+- refactor(sandbox): use centralized sandbox paths in macOS Seatbelt
+  implementation by @ehedlund in
+  [#24984](https://github.com/google-gemini/gemini-cli/pull/24984)
+- feat(cli): refine tool output formatting for compact mode by @jwhelangoog in
+  [#24677](https://github.com/google-gemini/gemini-cli/pull/24677)
+- fix(sdk): skip broken sendStream tests to unblock nightly by @SandyTao520 in
+  [#25000](https://github.com/google-gemini/gemini-cli/pull/25000)
+- refactor(core): use centralized path resolution for Linux sandbox by @ehedlund
+  in [#24985](https://github.com/google-gemini/gemini-cli/pull/24985)
+- Support ctrl+shift+g by @jacob314 in
+  [#25035](https://github.com/google-gemini/gemini-cli/pull/25035)
+- feat(core): refactor subagent tool to unified invoke_subagent tool by
+  @abhipatel12 in
+  [#24489](https://github.com/google-gemini/gemini-cli/pull/24489)
+- fix(core): add explicit git identity env vars to prevent sandbox checkpointing
+  error by @mrpmohiburrahman in
+  [#19775](https://github.com/google-gemini/gemini-cli/pull/19775)
+- fix: respect hideContextPercentage when FooterConfigDialog is closed without
+  changes by @chernistry in
+  [#24773](https://github.com/google-gemini/gemini-cli/pull/24773)
+- fix(cli): suppress unhandled AbortError logs during request cancellation by
+  @euxaristia in
+  [#22621](https://github.com/google-gemini/gemini-cli/pull/22621)
+- Automated documentation audit by @g-samroberts in
+  [#24567](https://github.com/google-gemini/gemini-cli/pull/24567)
+- feat(cli): implement useAgentStream hook by @mbleigh in
+  [#24292](https://github.com/google-gemini/gemini-cli/pull/24292)
+- refactor(plan) Clean default plan toml by @ruomengz in
+  [#25037](https://github.com/google-gemini/gemini-cli/pull/25037)
+- refactor(core): remove legacy subagent wrapping tools by @abhipatel12 in
+  [#25053](https://github.com/google-gemini/gemini-cli/pull/25053)
+- fix(core): honor retryDelay in RetryInfo for 503 errors by @yunaseoul in
+  [#25057](https://github.com/google-gemini/gemini-cli/pull/25057)
+- fix(core): remediate subagent memory leaks using AbortSignal in MessageBus by
+  @abhipatel12 in
+  [#25048](https://github.com/google-gemini/gemini-cli/pull/25048)
+- feat(cli): wire up useAgentStream in AppContainer by @mbleigh in
+  [#24297](https://github.com/google-gemini/gemini-cli/pull/24297)
+- feat(core): migrate chat recording to JSONL streaming by @spencer426 in
+  [#23749](https://github.com/google-gemini/gemini-cli/pull/23749)
+- fix(core): clear 5-minute timeouts in oauth flow to prevent memory leaks by
+  @spencer426 in
+  [#24968](https://github.com/google-gemini/gemini-cli/pull/24968)
+- fix(sandbox): centralize async git worktree resolution and enforce read-only
+  security by @ehedlund in
+  [#25040](https://github.com/google-gemini/gemini-cli/pull/25040)
+- feat(test): add high-volume shell test and refine perf harness by @sripasg in
+  [#24983](https://github.com/google-gemini/gemini-cli/pull/24983)
+- fix(core): silently handle EPERM when listing dir structure by @scidomino in
+  [#25066](https://github.com/google-gemini/gemini-cli/pull/25066)
+- Changelog for v0.37.1 by @gemini-cli-robot in
+  [#25055](https://github.com/google-gemini/gemini-cli/pull/25055)
+- fix: decode Uint8Array and multi-byte UTF-8 in API error messages by
+  @kimjune01 in [#23341](https://github.com/google-gemini/gemini-cli/pull/23341)
+- Automated documentation audit results by @g-samroberts in
+  [#22755](https://github.com/google-gemini/gemini-cli/pull/22755)
+- debugging(ui): add optional debugRainbow setting by @jacob314 in
+  [#25088](https://github.com/google-gemini/gemini-cli/pull/25088)
+- fix: resolve lifecycle memory leaks by cleaning up listeners and root closures
   by @spencer426 in
-  [#24566](https://github.com/google-gemini/gemini-cli/pull/24566)
-- Add an eval for and fix unsafe cloning behavior. by @gundermanc in
-  [#24457](https://github.com/google-gemini/gemini-cli/pull/24457)
-- fix(policy): allow complete_task in plan mode by @abhipatel12 in
-  [#24771](https://github.com/google-gemini/gemini-cli/pull/24771)
-- feat(telemetry): add browser agent clearcut metrics by @gsquared94 in
-  [#24688](https://github.com/google-gemini/gemini-cli/pull/24688)
-- feat(cli): support selective topic expansion and click-to-expand by
-  @Abhijit-2592 in
-  [#24793](https://github.com/google-gemini/gemini-cli/pull/24793)
-- temporarily disable sandbox integration test on windows by @ehedlund in
-  [#24786](https://github.com/google-gemini/gemini-cli/pull/24786)
-- Remove flakey test by @scidomino in
-  [#24837](https://github.com/google-gemini/gemini-cli/pull/24837)
-- Alisa/approve button by @alisa-alisa in
-  [#24645](https://github.com/google-gemini/gemini-cli/pull/24645)
-- feat(hooks): display hook system messages in UI by @mbleigh in
-  [#24616](https://github.com/google-gemini/gemini-cli/pull/24616)
-- fix(core): propagate BeforeModel hook model override end-to-end by @krishdef7
-  in [#24784](https://github.com/google-gemini/gemini-cli/pull/24784)
-- chore: fix formatting for behavioral eval skill reference file by @abhipatel12
-  in [#24846](https://github.com/google-gemini/gemini-cli/pull/24846)
-- fix: use directory junctions on Windows for skill linking by @enjoykumawat in
-  [#24823](https://github.com/google-gemini/gemini-cli/pull/24823)
-- fix(cli): prevent multiple banner increments on remount by @sehoon38 in
-  [#24843](https://github.com/google-gemini/gemini-cli/pull/24843)
-- feat(acp): add /help command by @sripasg in
-  [#24839](https://github.com/google-gemini/gemini-cli/pull/24839)
-- fix(core): remove tmux alternate buffer warning by @jackwotherspoon in
-  [#24852](https://github.com/google-gemini/gemini-cli/pull/24852)
-- Improve sandbox error matching and caching by @DavidAPierce in
-  [#24550](https://github.com/google-gemini/gemini-cli/pull/24550)
-- feat(core): add agent protocol UI types and experimental flag by @mbleigh in
-  [#24275](https://github.com/google-gemini/gemini-cli/pull/24275)
-- feat(core): use experiment flags for default fetch timeouts by @yunaseoul in
-  [#24261](https://github.com/google-gemini/gemini-cli/pull/24261)
-- Revert "fix(ui): improve narration suppression and reduce flicker (#2… by
+  [#25049](https://github.com/google-gemini/gemini-cli/pull/25049)
+- docs(cli): updates f12 description to be more precise by @JayadityaGit in
+  [#15816](https://github.com/google-gemini/gemini-cli/pull/15816)
+- fix(cli): mark /settings as unsafe to run concurrently by @jacob314 in
+  [#25061](https://github.com/google-gemini/gemini-cli/pull/25061)
+- fix(core): remove buffer slice to prevent OOM on large output streams by
+  @spencer426 in
+  [#25094](https://github.com/google-gemini/gemini-cli/pull/25094)
+- feat(core): persist subagent agentId in tool call records by @abhipatel12 in
+  [#25092](https://github.com/google-gemini/gemini-cli/pull/25092)
+- chore(core): increase codebase investigator turn limits to 50 by @abhipatel12
+  in [#25125](https://github.com/google-gemini/gemini-cli/pull/25125)
+- refactor(core): consolidate execute() arguments into ExecuteOptions by
+  @mbleigh in [#25101](https://github.com/google-gemini/gemini-cli/pull/25101)
+- feat(core): add Strategic Re-evaluation guidance to system prompt by
+  @aishaneeshah in
+  [#25062](https://github.com/google-gemini/gemini-cli/pull/25062)
+- fix(core): preserve shell execution config fields on update by
+  @jasonmatthewsuhari in
+  [#25113](https://github.com/google-gemini/gemini-cli/pull/25113)
+- docs: add vi shortcuts and clarify MCP sandbox setup by @chrisjcthomas in
+  [#21679](https://github.com/google-gemini/gemini-cli/pull/21679)
+- fix(cli): pass session id to interactive shell executions by
+  @jasonmatthewsuhari in
+  [#25114](https://github.com/google-gemini/gemini-cli/pull/25114)
+- fix(cli): resolve text sanitization data loss due to C1 control characters by
+  @euxaristia in
+  [#22624](https://github.com/google-gemini/gemini-cli/pull/22624)
+- feat(core): add large memory regression test by @cynthialong0-0 in
+  [#25059](https://github.com/google-gemini/gemini-cli/pull/25059)
+- fix(core): resolve PTY exhaustion and orphan MCP subprocess leaks by
+  @spencer426 in
+  [#25079](https://github.com/google-gemini/gemini-cli/pull/25079)
+- chore(deps): update vulnerable dependencies via npm audit fix by @scidomino in
+  [#25140](https://github.com/google-gemini/gemini-cli/pull/25140)
+- perf(sandbox): optimize Windows sandbox initialization via native ACL
+  application by @ehedlund in
+  [#25077](https://github.com/google-gemini/gemini-cli/pull/25077)
+- chore: switch from keytar to @github/keytar by @cocosheng-g in
+  [#25143](https://github.com/google-gemini/gemini-cli/pull/25143)
+- fix: improve audio MIME normalization and validation in file reads by
+  @junaiddshaukat in
+  [#21636](https://github.com/google-gemini/gemini-cli/pull/21636)
+- docs: Update docs-audit to include changes in PR body by @g-samroberts in
+  [#25153](https://github.com/google-gemini/gemini-cli/pull/25153)
+- docs: correct documentation for enforced authentication type by @cocosheng-g
+  in [#25142](https://github.com/google-gemini/gemini-cli/pull/25142)
+- fix(cli): exclude update_topic from confirmation queue count by @Abhijit-2592
+  in [#24945](https://github.com/google-gemini/gemini-cli/pull/24945)
+- Memory fix for trace's streamWrapper. by @anthraxmilkshake in
+  [#25089](https://github.com/google-gemini/gemini-cli/pull/25089)
+- fix(core): fix quota footer for non-auto models and improve display by
+  @jackwotherspoon in
+  [#25121](https://github.com/google-gemini/gemini-cli/pull/25121)
+- docs(contributing): clarify self-assignment policy for issues by @jmr in
+  [#23087](https://github.com/google-gemini/gemini-cli/pull/23087)
+- feat(core): add skill patching support with /memory inbox integration by
+  @SandyTao520 in
+  [#25148](https://github.com/google-gemini/gemini-cli/pull/25148)
+- Stop suppressing thoughts and text in model response by @gundermanc in
+  [#25073](https://github.com/google-gemini/gemini-cli/pull/25073)
+- fix(release): prefix git hash in nightly versions to prevent semver
+  normalization by @SandyTao520 in
+  [#25304](https://github.com/google-gemini/gemini-cli/pull/25304)
+- feat(cli): extract QuotaContext and resolve infinite render loop by @Adib234
+  in [#24959](https://github.com/google-gemini/gemini-cli/pull/24959)
+- refactor(core): extract and centralize sandbox path utilities by @ehedlund in
+  [#25305](https://github.com/google-gemini/gemini-cli/pull/25305)
+- feat(ui): added enhancements to scroll momentum by @devr0306 in
+  [#24447](https://github.com/google-gemini/gemini-cli/pull/24447)
+- fix(core): replace custom binary detection with isbinaryfile to correctly
+  handle UTF-8 (U+FFFD) by @Anjaligarhwal in
+  [#25297](https://github.com/google-gemini/gemini-cli/pull/25297)
+- feat(agent): implement tool-controlled display protocol (Steps 2-3) by
+  @mbleigh in [#25134](https://github.com/google-gemini/gemini-cli/pull/25134)
+- Stop showing scrollbar unless we are in terminalBuffer mode by @jacob314 in
+  [#25320](https://github.com/google-gemini/gemini-cli/pull/25320)
+- feat: support auth block in MCP servers config in agents by @TanmayVartak in
+  [#24770](https://github.com/google-gemini/gemini-cli/pull/24770)
+- fix(core): expose GEMINI_PLANS_DIR to hook environment by @Adib234 in
+  [#25296](https://github.com/google-gemini/gemini-cli/pull/25296)
+- feat(core): implement silent fallback for Plan Mode model routing by @jerop in
+  [#25317](https://github.com/google-gemini/gemini-cli/pull/25317)
+- fix: correct redirect count increment in fetchJson by @KevinZhao in
+  [#24896](https://github.com/google-gemini/gemini-cli/pull/24896)
+- fix(core): prevent secondary crash in ModelRouterService finally block by
   @gundermanc in
-  [#24857](https://github.com/google-gemini/gemini-cli/pull/24857)
-- refactor(cli): remove duplication in interactive shell awaiting input hint by
-  @JayadityaGit in
-  [#24801](https://github.com/google-gemini/gemini-cli/pull/24801)
-- refactor(core): make LegacyAgentSession dependencies optional by @mbleigh in
-  [#24287](https://github.com/google-gemini/gemini-cli/pull/24287)
-- Changelog for v0.37.0-preview.2 by @gemini-cli-robot in
-  [#24848](https://github.com/google-gemini/gemini-cli/pull/24848)
-- fix(cli): always show shell command description or actual command by @jacob314
-  in [#24774](https://github.com/google-gemini/gemini-cli/pull/24774)
-- Added flag for ept size and increased default size by @devr0306 in
-  [#24859](https://github.com/google-gemini/gemini-cli/pull/24859)
-- fix(core): dispose Scheduler to prevent McpProgress listener leak by
-  @Anjaligarhwal in
-  [#24870](https://github.com/google-gemini/gemini-cli/pull/24870)
-- fix(cli): switch default back to terminalBuffer=false and fix regressions
-  introduced for that mode by @jacob314 in
-  [#24873](https://github.com/google-gemini/gemini-cli/pull/24873)
-- feat(cli): switch to ctrl+g from ctrl-x by @jacob314 in
-  [#24861](https://github.com/google-gemini/gemini-cli/pull/24861)
-- fix: isolate concurrent browser agent instances by @gsquared94 in
-  [#24794](https://github.com/google-gemini/gemini-cli/pull/24794)
-- docs: update MCP server OAuth redirect port documentation by @adamfweidman in
-  [#24844](https://github.com/google-gemini/gemini-cli/pull/24844)
+  [#25333](https://github.com/google-gemini/gemini-cli/pull/25333)
+- feat(core): introduce decoupled ContextManager and Sidecar architecture by
+  @joshualitt in
+  [#24752](https://github.com/google-gemini/gemini-cli/pull/24752)
+- docs(core): update generalist agent documentation by @abhipatel12 in
+  [#25325](https://github.com/google-gemini/gemini-cli/pull/25325)
+- chore(mcp): check MCP error code over brittle string match by @jackwotherspoon
+  in [#25381](https://github.com/google-gemini/gemini-cli/pull/25381)
+- feat(plan): update plan mode prompt to allow showing plan content by @ruomengz
+  in [#25058](https://github.com/google-gemini/gemini-cli/pull/25058)
+- test(core): improve sandbox integration test coverage and fix OS-specific
+  failures by @ehedlund in
+  [#25307](https://github.com/google-gemini/gemini-cli/pull/25307)
+- fix(core): use debug level for keychain fallback logging by @ehedlund in
+  [#25398](https://github.com/google-gemini/gemini-cli/pull/25398)
+- feat(test): add a performance test in asian language by @cynthialong0-0 in
+  [#25392](https://github.com/google-gemini/gemini-cli/pull/25392)
+- feat(cli): enable mouse clicking for cursor positioning in AskUser multi-line
+  answers by @Adib234 in
+  [#24630](https://github.com/google-gemini/gemini-cli/pull/24630)
+- fix(core): detect kmscon terminal as supporting true color by @claygeo in
+  [#25282](https://github.com/google-gemini/gemini-cli/pull/25282)
+- ci: add agent session drift check workflow by @adamfweidman in
+  [#25389](https://github.com/google-gemini/gemini-cli/pull/25389)
+- use macos-latest-large runner where applicable. by @scidomino in
+  [#25413](https://github.com/google-gemini/gemini-cli/pull/25413)
+- Changelog for v0.37.2 by @gemini-cli-robot in
+  [#25336](https://github.com/google-gemini/gemini-cli/pull/25336)
+- fix(patch): cherry-pick a4e98c0 to release/v0.39.0-preview.0-pr-25138 to patch
+  version v0.39.0-preview.0 and create version 0.39.0-preview.1 by
+  @gemini-cli-robot in
+  [#25766](https://github.com/google-gemini/gemini-cli/pull/25766)
+- fix(patch): cherry-pick d6f88f8 to release/v0.39.0-preview.1-pr-25670 to patch
+  version v0.39.0-preview.1 and create version 0.39.0-preview.2 by
+  @gemini-cli-robot in
+  [#25776](https://github.com/google-gemini/gemini-cli/pull/25776)
 
 **Full Changelog**:
-https://github.com/google-gemini/gemini-cli/compare/v0.38.0...v0.38.2
+https://github.com/google-gemini/gemini-cli/compare/v0.38.2...v0.39.0

docs/changelogs/preview.md

@@ -1,6 +1,6 @@
-# Preview release: v0.39.0-preview.0
+# Preview release: v0.40.0-preview.3
 
-Released: April 14, 2026
+Released: April 24, 2026
 
 Our preview release includes the latest, new, and experimental features. This
 release may not be as stable as our [latest weekly release](latest.md).
@@ -13,24 +13,174 @@ npm install -g @google/gemini-cli@preview
 
 ## Highlights
 
-- **Refactored Subagents and Unified Tooling:** Consolidate subagent tools into
-  a single `invoke_subagent` tool, removed legacy wrapping tools, and improved
-  turn limits for codebase investigator.
-- **Advanced Memory and Skill Management:** Introduced `/memory` inbox for
-  reviewing extracted skills and added skill patching support, enhancing agent
-  learning and persistence.
-- **Expanded Test and Evaluation Infrastructure:** Added memory and CPU
-  performance integration test harnesses and generalized evaluation
-  infrastructure for better suite organization.
-- **Sandbox and Security Hardening:** Centralized sandbox paths for Linux and
-  macOS, enforced read-only security for async git worktree resolution, and
-  optimized Windows sandbox initialization.
-- **Enhanced CLI UX and UI Stability:** Improved scroll momentum, added a
-  `debugRainbow` setting, and resolved various memory leaks and PTY exhaustion
-  issues for a smoother terminal experience.
+- **Ripgrep Binary Bundling:** Ripgrep binaries are now bundled into the Single
+  Executable Application (SEA), enabling grep functionality in offline
+  environments.
+- **MCP Resource Tools:** New core tools added to list and read MCP (Model
+  Context Protocol) resources, expanding the agent's ability to interact with
+  MCP servers.
+- **Local Model Setup:** Introduced a streamlined `gemini gemma` command for
+  easier local model setup and integration.
+- **Prompt-Driven Memory Management:** Refactored memory management into a
+  prompt-driven, four-tier system and integrated `skill-creator` for robust
+  skill extraction.
+- **Enhanced UI and Accessibility:** Added support for OSC 777 terminal
+  notifications and GitHub colorblind themes for better user feedback and
+  accessibility.
 
 ## What's Changed
 
+- update FatalUntrustedWorkspaceError message to include doc link by @ehedlund
+  in [#25874](https://github.com/google-gemini/gemini-cli/pull/25874)
+- feat(core): enhance shell command validation and add core tools allowlist by
+  @galz10 in [#25720](https://github.com/google-gemini/gemini-cli/pull/25720)
+- feat(cli): secure .env loading and enforce workspace trust in headless mode by
+  @ehedlund in [#25814](https://github.com/google-gemini/gemini-cli/pull/25814)
+- chore(release): bump version to 0.40.0-nightly.20260414.g5b1f7375a by
+  @gemini-cli-robot in
+  [#25420](https://github.com/google-gemini/gemini-cli/pull/25420)
+- Fix(core): retry additional OpenSSL 3.x SSL errors during streaming (#16075)
+  by @rcleveng in
+  [#25187](https://github.com/google-gemini/gemini-cli/pull/25187)
+- fix(core): prevent YOLO mode from being downgraded by @galz10 in
+  [#25341](https://github.com/google-gemini/gemini-cli/pull/25341)
+- feat: bundle ripgrep binaries into SEA for offline support by @scidomino in
+  [#25342](https://github.com/google-gemini/gemini-cli/pull/25342)
+- Changelog for v0.39.0-preview.0 by @gemini-cli-robot in
+  [#25417](https://github.com/google-gemini/gemini-cli/pull/25417)
+- feat(test): add large conversation scenario for performance test by
+  @cynthialong0-0 in
+  [#25331](https://github.com/google-gemini/gemini-cli/pull/25331)
+- improve(core): require recurrence evidence before extracting skills by
+  @SandyTao520 in
+  [#25147](https://github.com/google-gemini/gemini-cli/pull/25147)
+- test(evals): add subagent delegation evaluation tests by @anj-s in
+  [#24619](https://github.com/google-gemini/gemini-cli/pull/24619)
+- feat: add github colorblind themes by @Z1xus in
+  [#15504](https://github.com/google-gemini/gemini-cli/pull/15504)
+- fix(core): honor GOOGLE_GEMINI_BASE_URL and GOOGLE_VERTEX_BASE_URL by
+  @chrisjcthomas in
+  [#25357](https://github.com/google-gemini/gemini-cli/pull/25357)
+- fix(cli): clean up slash command IDE listeners by @jasonmatthewsuhari in
+  [#24397](https://github.com/google-gemini/gemini-cli/pull/24397)
+- Changelog for v0.38.0 by @gemini-cli-robot in
+  [#25470](https://github.com/google-gemini/gemini-cli/pull/25470)
+- fix(evals): update eval tests for invoke_agent telemetry and project-scoped
+  memory by @SandyTao520 in
+  [#25502](https://github.com/google-gemini/gemini-cli/pull/25502)
+- Changelog for v0.38.1 by @gemini-cli-robot in
+  [#25476](https://github.com/google-gemini/gemini-cli/pull/25476)
+- feat(core): integrate skill-creator into skill extraction agent by
+  @SandyTao520 in
+  [#25421](https://github.com/google-gemini/gemini-cli/pull/25421)
+- feat(cli): provide default post-submit prompt for skill command by @ruomengz
+  in [#25327](https://github.com/google-gemini/gemini-cli/pull/25327)
+- feat(core): add tools to list and read MCP resources by @ruomengz in
+  [#25395](https://github.com/google-gemini/gemini-cli/pull/25395)
+- fix(evals): add typecheck coverage for evals, integration-tests, and
+  memory-tests by @SandyTao520 in
+  [#25480](https://github.com/google-gemini/gemini-cli/pull/25480)
+- Use OSC 777 for terminal notifications by @jackyliuxx in
+  [#25300](https://github.com/google-gemini/gemini-cli/pull/25300)
+- fix(extensions): fix bundling for examples by @abhipatel12 in
+  [#25542](https://github.com/google-gemini/gemini-cli/pull/25542)
+- fix(cli): reset plan session state on /clear by @jasonmatthewsuhari in
+  [#25515](https://github.com/google-gemini/gemini-cli/pull/25515)
+- feat(core): add .mdx support to get-internal-docs tool by @g-samroberts in
+  [#25090](https://github.com/google-gemini/gemini-cli/pull/25090)
+- docs(policy): mention that workspace policies are broken by @6112 in
+  [#24367](https://github.com/google-gemini/gemini-cli/pull/24367)
+- fix(core): allow explicit write permissions to override governance file
+  protections in sandboxes by @galz10 in
+  [#25338](https://github.com/google-gemini/gemini-cli/pull/25338)
+- feat(sandbox): resolve custom seatbelt profiles from $HOME/.gemini first by
+  @mvanhorn in [#25427](https://github.com/google-gemini/gemini-cli/pull/25427)
+- Reduce blank lines. by @gundermanc in
+  [#25563](https://github.com/google-gemini/gemini-cli/pull/25563)
+- fix(ui): revert preview theme on dialog unmount by @JayadityaGit in
+  [#22542](https://github.com/google-gemini/gemini-cli/pull/22542)
+- fix(core): fix ShellExecutionConfig spread and add ProjectRegistry save
+  backoff by @mahimashanware in
+  [#25382](https://github.com/google-gemini/gemini-cli/pull/25382)
+- feat(core): Disable topic updates for subagents by @gundermanc in
+  [#25567](https://github.com/google-gemini/gemini-cli/pull/25567)
+- feat(core): enable topic update narration by default and promote to general by
+  @gundermanc in
+  [#25586](https://github.com/google-gemini/gemini-cli/pull/25586)
+- docs: migrate installation and authentication to mdx with tabbed layouts by
+  @g-samroberts in
+  [#25155](https://github.com/google-gemini/gemini-cli/pull/25155)
+- feat(config): split memoryManager flag into autoMemory by @SandyTao520 in
+  [#25601](https://github.com/google-gemini/gemini-cli/pull/25601)
+- fix(core): allow Cloud Shell users to use PRO_MODEL_NO_ACCESS experiment by
+  @sehoon38 in [#25702](https://github.com/google-gemini/gemini-cli/pull/25702)
+- fix(cli): round slow render latency to avoid opentelemetry float warning by
+  @scidomino in [#25709](https://github.com/google-gemini/gemini-cli/pull/25709)
+- docs(tracker): introduce experimental task tracker feature by @anj-s in
+  [#24556](https://github.com/google-gemini/gemini-cli/pull/24556)
+- docs(cli): fix inconsistent system.md casing in system prompt docs by @Bodlux
+  in [#25414](https://github.com/google-gemini/gemini-cli/pull/25414)
+- feat(cli): add streamlined `gemini gemma` local model setup by @Samee24 in
+  [#25498](https://github.com/google-gemini/gemini-cli/pull/25498)
+- Changelog for v0.38.2 by @gemini-cli-robot in
+  [#25593](https://github.com/google-gemini/gemini-cli/pull/25593)
+- Fix: Disallow overriding IDE stdio via workspace .env (RCE) by @M0nd0R in
+  [#25022](https://github.com/google-gemini/gemini-cli/pull/25022)
+- feat(test): refactor the memory usage test to use metrics from CLI process
+  instead of test runner by @cynthialong0-0 in
+  [#25708](https://github.com/google-gemini/gemini-cli/pull/25708)
+- feat(vertex): add settings for Vertex AI request routing by @gordonhwc in
+  [#25513](https://github.com/google-gemini/gemini-cli/pull/25513)
+- Fix/allow for session persistence by @ahsanfarooq210 in
+  [#25176](https://github.com/google-gemini/gemini-cli/pull/25176)
+- Allow dots on GEMINI_API_KEY by @DKbyo in
+  [#25497](https://github.com/google-gemini/gemini-cli/pull/25497)
+- feat(telemetry): add flag for enabling traces specifically by @spencer426 in
+  [#25343](https://github.com/google-gemini/gemini-cli/pull/25343)
+- fix(core): resolve nested plan directory duplication and relative path
+  policies by @mahimashanware in
+  [#25138](https://github.com/google-gemini/gemini-cli/pull/25138)
+- feat: detect new files in @ recommendations with watcher based updates by
+  @prassamin in [#25256](https://github.com/google-gemini/gemini-cli/pull/25256)
+- fix(cli): use newline in shell command wrapping to avoid breaking heredocs by
+  @cocosheng-g in
+  [#25537](https://github.com/google-gemini/gemini-cli/pull/25537)
+- fix(cli): ensure theme dialog labels are rendered for all themes by
+  @JayadityaGit in
+  [#24599](https://github.com/google-gemini/gemini-cli/pull/24599)
+- fix(core): disable detached mode in Bun to prevent immediate SIGHUP of child
+  processes by @euxaristia in
+  [#22620](https://github.com/google-gemini/gemini-cli/pull/22620)
+- feat: add /new as alias for /clear and refine command description by @ved015
+  in [#17865](https://github.com/google-gemini/gemini-cli/pull/17865)
+- fix(cli): start auto memory in ACP sessions by @jasonmatthewsuhari in
+  [#25626](https://github.com/google-gemini/gemini-cli/pull/25626)
+- fix(core): remove duplicate initialize call on agents refreshed by
+  @adamfweidman in
+  [#25670](https://github.com/google-gemini/gemini-cli/pull/25670)
+- test(e2e): default integration tests to Flash Preview by @SandyTao520 in
+  [#25753](https://github.com/google-gemini/gemini-cli/pull/25753)
+- refactor(memory): replace MemoryManagerAgent with prompt-driven memory editing
+  across four tiers by @SandyTao520 in
+  [#25716](https://github.com/google-gemini/gemini-cli/pull/25716)
+- fix(cli): fix "/clear (new)" command by @mini2s in
+  [#25801](https://github.com/google-gemini/gemini-cli/pull/25801)
+- fix(core): use dynamic CLI version for IDE client instead of hardcoded '1.0.0'
+  by @thekishandev in
+  [#24414](https://github.com/google-gemini/gemini-cli/pull/24414)
+- fix(core): handle line endings in ignore file parsing by @xoma-zver in
+  [#23895](https://github.com/google-gemini/gemini-cli/pull/23895)
+- Fix/command injection shell by @Famous077 in
+  [#24170](https://github.com/google-gemini/gemini-cli/pull/24170)
+- fix(ui): removed background color for input by @devr0306 in
+  [#25339](https://github.com/google-gemini/gemini-cli/pull/25339)
+- fix(devtools): reduce memory usage and defer connection by @SandyTao520 in
+  [#24496](https://github.com/google-gemini/gemini-cli/pull/24496)
+- fix(core): support jsonl session logs in memory and summary services by
+  @SandyTao520 in
+  [#25816](https://github.com/google-gemini/gemini-cli/pull/25816)
+- fix(release): exclude ripgrep binaries from npm tarballs by @SandyTao520 in
+  [#25841](https://github.com/google-gemini/gemini-cli/pull/25841)
 - refactor(plan): simplify policy priorities and consolidate read-only rules by
   @ruomengz in [#24849](https://github.com/google-gemini/gemini-cli/pull/24849)
 - feat(test-utils): add memory usage integration test harness by @sripasg in
@@ -254,4 +404,4 @@ npm install -g @google/gemini-cli@preview
   [#25336](https://github.com/google-gemini/gemini-cli/pull/25336)
 
 **Full Changelog**:
-https://github.com/google-gemini/gemini-cli/compare/v0.38.0-preview.0...v0.39.0-preview.0
+https://github.com/google-gemini/gemini-cli/compare/v0.38.0-preview.0...v0.40.0-preview.3

docs/cli/plan-mode.md

@@ -470,7 +470,8 @@ associated plan files and task trackers.
 
 - **Default behavior:** Sessions (and their plans) are retained for **30 days**.
 - **Configuration:** You can customize this behavior via the `/settings` command
-  (search for **Session Retention**) or in your `settings.json` file. See
+  (search for **Enable Session Cleanup** or **Keep chat history**) or in your
+  `settings.json` file. See
   [session retention](../cli/session-management.md#session-retention) for more
   details.
 

docs/cli/sandbox.md

@@ -31,6 +31,53 @@ The benefits of sandboxing include:
 - **Safety**: Reduce risk when working with untrusted code or experimental
   commands.
 
+## Quickstart
+
+You can enable sandboxing using a command flag, environment variable, or
+configuration file.
+
+### Using the command flag
+
+```bash
+gemini -s -p "analyze the code structure"
+```
+
+### Using an environment variable
+
+**macOS/Linux**
+
+```bash
+export GEMINI_SANDBOX=true
+gemini -p "run the test suite"
+```
+
+**Windows (PowerShell)**
+
+```powershell
+$env:GEMINI_SANDBOX="true"
+gemini -p "run the test suite"
+```
+
+### Configuring via settings.json
+
+```json
+{
+  "tools": {
+    "sandbox": "docker"
+  }
+}
+```
+
+## Configuration
+
+Enable sandboxing using one of the following methods (in order of precedence):
+
+1. **Command flag**: `-s` or `--sandbox`
+2. **Environment variable**:
+   `GEMINI_SANDBOX=true|docker|podman|sandbox-exec|runsc|lxc`
+3. **Settings file**: `"sandbox": true` in the `tools` object of your
+   `settings.json` file (for example, `{"tools": {"sandbox": true}}`).
+
 ## Sandboxing methods
 
 Your ideal method of sandboxing may differ depending on your platform and your
@@ -43,12 +90,92 @@ Lightweight, built-in sandboxing using `sandbox-exec`.
 **Default profile**: `permissive-open` - restricts writes outside project
 directory but allows most other operations.
 
+Built-in profiles (set via `SEATBELT_PROFILE` env var):
+
+- `permissive-open` (default): Write restrictions, network allowed
+- `permissive-proxied`: Write restrictions, network via proxy
+- `restrictive-open`: Strict restrictions, network allowed
+- `restrictive-proxied`: Strict restrictions, network via proxy
+- `strict-open`: Read and write restrictions, network allowed
+- `strict-proxied`: Read and write restrictions, network via proxy
+
 ### 2. Container-based (Docker/Podman)
 
-Cross-platform sandboxing with complete process isolation.
+Cross-platform sandboxing with complete process isolation using container
+technology. By default, it uses the `ghcr.io/google/gemini-cli:latest` image.
 
-**Note**: Requires building the sandbox image locally or using a published image
-from your organization's registry.
+**Prerequisites:**
+
+- Docker or Podman must be installed and running on your system.
+
+**How it works (Workspace directory):**
+
+Inside the sandbox container, your current working directory is mounted at the
+**exact same absolute path** as it is on your host machine. For example, if you
+run the CLI from `/Users/you/project` on your host machine, the sandbox will
+mount your local project folder and operate within `/Users/you/project` inside
+the container. This allows the AI to seamlessly read and modify your project
+files while remaining isolated from the rest of your system.
+
+**Quick setup:**
+
+To enable Docker sandboxing, run Gemini CLI with the sandbox flag and specify
+Docker as the provider:
+
+```bash
+# Using the environment variable (Recommended)
+export GEMINI_SANDBOX=docker
+gemini -p "build the project"
+
+# Or configure it permanently in your settings.json
+# {"tools": {"sandbox": "docker"}}
+```
+
+**Customizing the Sandbox Image:**
+
+If your project requires specific dependencies, you can specify a custom image
+name or have Gemini CLI build one for you automatically. You can use any Docker
+or Podman image as your sandbox, provided it has standard shell utilities (like
+`bash`) available.
+
+**Option A: Using an existing custom image (e.g., Artifact Registry)**
+
+To configure a custom image that is hosted on a registry (or built locally),
+update your `settings.json` to use an object for the sandbox configuration, or
+set the `GEMINI_SANDBOX_IMAGE` environment variable.
+
+_Example: Configuring via `settings.json`_
+
+```json
+{
+  "tools": {
+    "sandbox": {
+      "command": "docker",
+      "image": "us-central1-docker.pkg.dev/my-project/my-repo/my-custom-sandbox:latest"
+    }
+  }
+}
+```
+
+_Example: Configuring via environment variable_
+
+```bash
+export GEMINI_SANDBOX_IMAGE="us-central1-docker.pkg.dev/my-project/my-repo/my-custom-sandbox:latest"
+```
+
+**Option B: Building a local custom image automatically**
+
+If you prefer to define your environment as code, you can provide a Dockerfile
+and Gemini CLI will build the image automatically.
+
+1.  Create a `.gemini/sandbox.Dockerfile` in your project root.
+2.  Ensure you have the `gh` CLI installed and authenticated (if you are using
+    the default `ghcr.io/google/gemini-cli` image as a base).
+3.  Run your command with the `BUILD_SANDBOX` environment variable set:
+
+```bash
+BUILD_SANDBOX=1 GEMINI_SANDBOX=docker gemini -p "run my custom build"
+```
 
 ### 3. Windows Native Sandbox (Windows only)
 
@@ -188,59 +315,49 @@ This mechanism ensures you don't have to manually re-run commands with more
 permissive sandbox settings, while still maintaining control over what the AI
 can access.
 
-## Quickstart
+### Including files outside the workspace
+
+By default, the sandbox only has access to the current project workspace. If you
+need the sandbox to have permission to operate on certain files or directories
+from the local file system outside of the project workspace, you can mount them
+using the `SANDBOX_MOUNTS` environment variable.
+
+Provide a comma-separated list of mount definitions in the format
+`from:to:opts`. If `to` is omitted, it defaults to the same path as `from`. If
+`opts` is omitted, it defaults to `ro` (read-only). Note that the `from` path
+must be an absolute path.
+
+**Example**:
 
 ```bash
-# Enable sandboxing with command flag
-gemini -s -p "analyze the code structure"
+export SANDBOX_MOUNTS="/path/on/host:/path/in/container:rw,/another/path:ro"
 ```
 
-**Use environment variable**
+## Running inside a Docker container
 
-**macOS/Linux**
+If you are running Gemini CLI itself from within an official or custom Docker
+container and want to enable sandboxing, you must share the host's Docker socket
+and ensure your workspace paths align.
+
+1.  **Mount the Docker socket**: Map `/var/run/docker.sock` so the CLI can spawn
+    sibling sandbox containers via the host's Docker daemon.
+2.  **Align workspace paths**: The path to your workspace inside the container
+    must exactly match the absolute path on the host. Because the sandbox
+    container is spawned by the host's Docker daemon, it resolves volume mounts
+    against the host file system.
+
+**Example**:
 
 ```bash
-export GEMINI_SANDBOX=true
-gemini -p "run the test suite"
+docker run -it \
+  -v /var/run/docker.sock:/var/run/docker.sock \
+  -v /absolute/path/on/host/project:/absolute/path/on/host/project \
+  -w /absolute/path/on/host/project \
+  -e GEMINI_SANDBOX=docker \
+  ghcr.io/google/gemini-cli:latest
 ```
 
-**Windows (PowerShell)**
-
-```powershell
-$env:GEMINI_SANDBOX="true"
-gemini -p "run the test suite"
-```
-
-**Configure in settings.json**
-
-```json
-{
-  "tools": {
-    "sandbox": "docker"
-  }
-}
-```
-
-## Configuration
-
-### Enable sandboxing (in order of precedence)
-
-1. **Command flag**: `-s` or `--sandbox`
-2. **Environment variable**:
-   `GEMINI_SANDBOX=true|docker|podman|sandbox-exec|runsc|lxc`
-3. **Settings file**: `"sandbox": true` in the `tools` object of your
-   `settings.json` file (for example, `{"tools": {"sandbox": true}}`).
-
-### macOS Seatbelt profiles
-
-Built-in profiles (set via `SEATBELT_PROFILE` env var):
-
-- `permissive-open` (default): Write restrictions, network allowed
-- `permissive-proxied`: Write restrictions, network via proxy
-- `restrictive-open`: Strict restrictions, network allowed
-- `restrictive-proxied`: Strict restrictions, network via proxy
-- `strict-open`: Read and write restrictions, network allowed
-- `strict-proxied`: Read and write restrictions, network via proxy
+## Advanced settings
 
 ### Custom sandbox flags
 
@@ -279,7 +396,7 @@ export SANDBOX_FLAGS="--flag1 --flag2=value"
 $env:SANDBOX_FLAGS="--flag1 --flag2=value"
 ```
 
-## Linux UID/GID handling
+### Linux UID/GID handling
 
 The sandbox automatically handles user permissions on Linux. Override these
 permissions with:

docs/cli/settings.md

@@ -161,20 +161,25 @@ they appear in the UI.
 
 ### Experimental
 
-| UI Label                                             | Setting                                         | Description                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                          | Default |
-| ---------------------------------------------------- | ----------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------- |
-| Gemma Models                                         | `experimental.gemma`                            | Enable access to Gemma 4 models (experimental).                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                      | `false` |
-| Enable Git Worktrees                                 | `experimental.worktrees`                        | Enable automated Git worktree management for parallel work.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                          | `false` |
-| Use OSC 52 Paste                                     | `experimental.useOSC52Paste`                    | Use OSC 52 for pasting. This may be more robust than the default system when using remote terminal sessions (if your terminal is configured to allow it).                                                                                                                                                                                                                                                                                                                                                                                                                                                            | `false` |
-| Use OSC 52 Copy                                      | `experimental.useOSC52Copy`                     | Use OSC 52 for copying. This may be more robust than the default system when using remote terminal sessions (if your terminal is configured to allow it).                                                                                                                                                                                                                                                                                                                                                                                                                                                            | `false` |
-| Model Steering                                       | `experimental.modelSteering`                    | Enable model steering (user hints) to guide the model during tool execution.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                         | `false` |
-| Direct Web Fetch                                     | `experimental.directWebFetch`                   | Enable web fetch behavior that bypasses LLM summarization.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                           | `false` |
-| Enable Gemma Model Router                            | `experimental.gemmaModelRouter.enabled`         | Enable the Gemma Model Router (experimental). Requires a local endpoint serving Gemma via the Gemini API using LiteRT-LM shim.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                       | `false` |
-| Auto-start LiteRT Server                             | `experimental.gemmaModelRouter.autoStartServer` | Automatically start the LiteRT-LM server when Gemini CLI starts and the Gemma router is enabled.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                     | `false` |
-| Memory v2                                            | `experimental.memoryV2`                         | Disable the built-in save_memory tool and let the main agent persist project context by editing markdown files directly with edit/write_file. Route facts across four tiers: team-shared conventions go to project GEMINI.md files, project-specific personal notes go to the per-project private memory folder (MEMORY.md as index + sibling .md files for detail), and cross-project personal preferences go to the global ~/.gemini/GEMINI.md (the only file under ~/.gemini/ that the agent can edit — settings, credentials, etc. remain off-limits). Set to false to fall back to the legacy save_memory tool. | `true`  |
-| Auto Memory                                          | `experimental.autoMemory`                       | Automatically extract reusable skills from past sessions in the background. Review results with /memory inbox.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                       | `false` |
-| Use the generalist profile to manage agent contexts. | `experimental.generalistProfile`                | Suitable for general coding and software development tasks.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                          | `false` |
-| Enable Context Management                            | `experimental.contextManagement`                | Enable logic for context management.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                 | `false` |
+| UI Label                                             | Setting                                         | Description                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                          | Default              |
+| ---------------------------------------------------- | ----------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------- |
+| Gemma Models                                         | `experimental.gemma`                            | Enable access to Gemma 4 models (experimental).                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                      | `false`              |
+| Voice Mode                                           | `experimental.voiceMode`                        | Enable experimental voice dictation and commands (/voice, /voice model).                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                             | `false`              |
+| Voice Activation Mode                                | `experimental.voice.activationMode`             | How to trigger voice recording with the Space key.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                   | `"push-to-talk"`     |
+| Voice Transcription Backend                          | `experimental.voice.backend`                    | The backend to use for voice transcription.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                          | `"gemini-live"`      |
+| Whisper Model                                        | `experimental.voice.whisperModel`               | The Whisper model to use for local transcription.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                    | `"ggml-base.en.bin"` |
+| Voice Stop Grace Period (ms)                         | `experimental.voice.stopGracePeriodMs`          | How long to wait for final transcription after stopping recording.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                   | `1000`               |
+| Enable Git Worktrees                                 | `experimental.worktrees`                        | Enable automated Git worktree management for parallel work.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                          | `false`              |
+| Use OSC 52 Paste                                     | `experimental.useOSC52Paste`                    | Use OSC 52 for pasting. This may be more robust than the default system when using remote terminal sessions (if your terminal is configured to allow it).                                                                                                                                                                                                                                                                                                                                                                                                                                                            | `false`              |
+| Use OSC 52 Copy                                      | `experimental.useOSC52Copy`                     | Use OSC 52 for copying. This may be more robust than the default system when using remote terminal sessions (if your terminal is configured to allow it).                                                                                                                                                                                                                                                                                                                                                                                                                                                            | `false`              |
+| Model Steering                                       | `experimental.modelSteering`                    | Enable model steering (user hints) to guide the model during tool execution.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                         | `false`              |
+| Direct Web Fetch                                     | `experimental.directWebFetch`                   | Enable web fetch behavior that bypasses LLM summarization.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                           | `false`              |
+| Enable Gemma Model Router                            | `experimental.gemmaModelRouter.enabled`         | Enable the Gemma Model Router (experimental). Requires a local endpoint serving Gemma via the Gemini API using LiteRT-LM shim.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                       | `false`              |
+| Auto-start LiteRT Server                             | `experimental.gemmaModelRouter.autoStartServer` | Automatically start the LiteRT-LM server when Gemini CLI starts and the Gemma router is enabled.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                     | `false`              |
+| Memory v2                                            | `experimental.memoryV2`                         | Disable the built-in save_memory tool and let the main agent persist project context by editing markdown files directly with edit/write_file. Route facts across four tiers: team-shared conventions go to project GEMINI.md files, project-specific personal notes go to the per-project private memory folder (MEMORY.md as index + sibling .md files for detail), and cross-project personal preferences go to the global ~/.gemini/GEMINI.md (the only file under ~/.gemini/ that the agent can edit — settings, credentials, etc. remain off-limits). Set to false to fall back to the legacy save_memory tool. | `true`               |
+| Auto Memory                                          | `experimental.autoMemory`                       | Automatically extract reusable skills from past sessions in the background. Review results with /memory inbox.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                       | `false`              |
+| Use the generalist profile to manage agent contexts. | `experimental.generalistProfile`                | Suitable for general coding and software development tasks.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                          | `false`              |
+| Enable Context Management                            | `experimental.contextManagement`                | Enable logic for context management.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                 | `false`              |
 
 ### Skills
 

docs/cli/tutorials/session-management.md

@@ -61,6 +61,19 @@ gemini --list-sessions
 gemini --delete-session 1
 ```
 
+### Scenario: Delete session on exit
+
+If you're doing a one-off task and don't want to leave any session history
+behind, use the `--delete` flag when exiting:
+
+```
+/exit --delete
+```
+
+This removes the current session's conversation history and tool output files
+before exiting. It's useful for privacy-sensitive tasks or quick one-off
+interactions.
+
 ## How to rewind time (Undo mistakes)
 
 Gemini CLI's **Rewind** feature is like `Ctrl+Z` for your workflow.

docs/reference/commands.md

@@ -323,6 +323,11 @@ Slash commands provide meta-level control over the CLI itself.
 ### `/quit` (or `/exit`)
 
 - **Description:** Exit Gemini CLI.
+- **Flags:**
+  - **`--delete`** _(optional)_: Exit and permanently delete the current
+    session's history and temporary files (chat recording, tool outputs). Useful
+    for privacy or one-off tasks where you don't want to leave any traces.
+  - **Usage:** `/quit --delete` or `/exit --delete`
 
 ### `/restore`
 

docs/reference/configuration.md

@@ -1191,7 +1191,7 @@ their corresponding top-level category object in your `settings.json` file.
           },
           "stateTransitions": {
             "terminal": "terminal",
-            "transient": "terminal",
+            "transient": "sticky_retry",
             "not_found": "terminal",
             "unknown": "terminal"
           }
@@ -1207,7 +1207,7 @@ their corresponding top-level category object in your `settings.json` file.
           },
           "stateTransitions": {
             "terminal": "terminal",
-            "transient": "terminal",
+            "transient": "sticky_retry",
             "not_found": "terminal",
             "unknown": "terminal"
           }
@@ -1224,7 +1224,7 @@ their corresponding top-level category object in your `settings.json` file.
           },
           "stateTransitions": {
             "terminal": "terminal",
-            "transient": "terminal",
+            "transient": "sticky_retry",
             "not_found": "terminal",
             "unknown": "terminal"
           }
@@ -1240,7 +1240,7 @@ their corresponding top-level category object in your `settings.json` file.
           },
           "stateTransitions": {
             "terminal": "terminal",
-            "transient": "terminal",
+            "transient": "sticky_retry",
             "not_found": "terminal",
             "unknown": "terminal"
           }
@@ -1257,7 +1257,7 @@ their corresponding top-level category object in your `settings.json` file.
           },
           "stateTransitions": {
             "terminal": "terminal",
-            "transient": "terminal",
+            "transient": "sticky_retry",
             "not_found": "terminal",
             "unknown": "terminal"
           }
@@ -1272,7 +1272,7 @@ their corresponding top-level category object in your `settings.json` file.
           },
           "stateTransitions": {
             "terminal": "terminal",
-            "transient": "terminal",
+            "transient": "sticky_retry",
             "not_found": "terminal",
             "unknown": "terminal"
           }
@@ -1288,7 +1288,7 @@ their corresponding top-level category object in your `settings.json` file.
           },
           "stateTransitions": {
             "terminal": "terminal",
-            "transient": "terminal",
+            "transient": "sticky_retry",
             "not_found": "terminal",
             "unknown": "terminal"
           }
@@ -1691,6 +1691,32 @@ their corresponding top-level category object in your `settings.json` file.
   - **Default:** `false`
   - **Requires restart:** Yes
 
+- **`experimental.voiceMode`** (boolean):
+  - **Description:** Enable experimental voice dictation and commands (/voice,
+    /voice model).
+  - **Default:** `false`
+
+- **`experimental.voice.activationMode`** (enum):
+  - **Description:** How to trigger voice recording with the Space key.
+  - **Default:** `"push-to-talk"`
+  - **Values:** `"push-to-talk"`, `"toggle"`
+
+- **`experimental.voice.backend`** (enum):
+  - **Description:** The backend to use for voice transcription.
+  - **Default:** `"gemini-live"`
+  - **Values:** `"gemini-live"`, `"whisper"`
+
+- **`experimental.voice.whisperModel`** (enum):
+  - **Description:** The Whisper model to use for local transcription.
+  - **Default:** `"ggml-base.en.bin"`
+  - **Values:** `"ggml-tiny.en.bin"`, `"ggml-base.en.bin"`,
+    `"ggml-large-v3-turbo-q5_0.bin"`, `"ggml-large-v3-turbo-q8_0.bin"`
+
+- **`experimental.voice.stopGracePeriodMs`** (number):
+  - **Description:** How long to wait for final transcription after stopping
+    recording.
+  - **Default:** `1000`
+
 - **`experimental.adk.agentSessionNoninteractiveEnabled`** (boolean):
   - **Description:** Enable non-interactive agent sessions.
   - **Default:** `false`
@@ -1820,6 +1846,12 @@ their corresponding top-level category object in your `settings.json` file.
   - **Default:** `true`
   - **Requires restart:** Yes
 
+- **`experimental.stressTestProfile`** (boolean):
+  - **Description:** Significantly lowers token limits to force early garbage
+    collection and distillation for testing purposes.
+  - **Default:** `false`
+  - **Requires restart:** Yes
+
 - **`experimental.autoMemory`** (boolean):
   - **Description:** Automatically extract reusable skills from past sessions in
     the background. Review results with /memory inbox.

docs/reference/keyboard-shortcuts.md

@@ -115,6 +115,7 @@ available combinations.
 | `app.restart`                 | Restart the application.                                                                                                                           | `R`<br />`Shift+R` |
 | `app.suspend`                 | Suspend the CLI and move it to the background.                                                                                                     | `Ctrl+Z`           |
 | `app.showShellUnfocusWarning` | Show warning when trying to move focus away from shell input.                                                                                      | `Tab`              |
+| `app.voiceModePTT`            | Hold to speak in Voice Mode.                                                                                                                       | `Space`            |
 
 #### Background Shell Controls
 

evals/save_memory.eval.ts

@@ -5,12 +5,78 @@
  */
 
 import { describe, expect } from 'vitest';
+import fs from 'node:fs';
+import path from 'node:path';
+import {
+  loadConversationRecord,
+  SESSION_FILE_PREFIX,
+} from '@google/gemini-cli-core';
 import {
   evalTest,
   assertModelHasOutput,
   checkModelOutputContent,
 } from './test-helper.js';
 
+function findDir(base: string, name: string): string | null {
+  if (!fs.existsSync(base)) return null;
+  const files = fs.readdirSync(base);
+  for (const file of files) {
+    const fullPath = path.join(base, file);
+    if (fs.statSync(fullPath).isDirectory()) {
+      if (file === name) return fullPath;
+      const found = findDir(fullPath, name);
+      if (found) return found;
+    }
+  }
+  return null;
+}
+
+async function loadLatestSessionRecord(homeDir: string, sessionId: string) {
+  const chatsDir = findDir(path.join(homeDir, '.gemini'), 'chats');
+  if (!chatsDir) {
+    throw new Error('Could not find chats directory for eval session logs');
+  }
+
+  const candidates = fs
+    .readdirSync(chatsDir)
+    .filter(
+      (file) =>
+        file.startsWith(SESSION_FILE_PREFIX) &&
+        (file.endsWith('.json') || file.endsWith('.jsonl')),
+    );
+
+  const matchingRecords = [];
+  for (const file of candidates) {
+    const filePath = path.join(chatsDir, file);
+    const record = await loadConversationRecord(filePath);
+    if (record?.sessionId === sessionId) {
+      matchingRecords.push(record);
+    }
+  }
+
+  matchingRecords.sort(
+    (a, b) => Date.parse(b.lastUpdated) - Date.parse(a.lastUpdated),
+  );
+  return matchingRecords[0] ?? null;
+}
+
+async function waitForSessionScratchpad(
+  homeDir: string,
+  sessionId: string,
+  timeoutMs = 30000,
+) {
+  const deadline = Date.now() + timeoutMs;
+  while (Date.now() < deadline) {
+    const record = await loadLatestSessionRecord(homeDir, sessionId);
+    if (record?.memoryScratchpad) {
+      return record;
+    }
+    await new Promise((resolve) => setTimeout(resolve, 1000));
+  }
+
+  return loadLatestSessionRecord(homeDir, sessionId);
+}
+
 describe('save_memory', () => {
   const TEST_PREFIX = 'Save memory test: ';
   const rememberingFavoriteColor = "Agent remembers user's favorite color";
@@ -569,6 +635,103 @@ describe('save_memory', () => {
     },
   });
 
+  const memoryV2SessionScratchpad =
+    'Session summary persists memory scratchpad for memory-saving sessions';
+  evalTest('USUALLY_PASSES', {
+    suiteName: 'default',
+    suiteType: 'behavioral',
+    name: memoryV2SessionScratchpad,
+    sessionId: 'memory-scratchpad-eval',
+    params: {
+      settings: {
+        experimental: { memoryV2: true },
+      },
+    },
+    messages: [
+      {
+        id: 'msg-1',
+        type: 'user',
+        content: [
+          {
+            text: 'Across all my projects, I prefer Vitest over Jest for testing.',
+          },
+        ],
+        timestamp: '2026-01-01T00:00:00Z',
+      },
+      {
+        id: 'msg-2',
+        type: 'gemini',
+        content: [{ text: 'Noted. What else should I keep in mind?' }],
+        timestamp: '2026-01-01T00:00:05Z',
+      },
+      {
+        id: 'msg-3',
+        type: 'user',
+        content: [
+          {
+            text: 'For this repo I was debugging a flaky API test earlier, but that was just transient context.',
+          },
+        ],
+        timestamp: '2026-01-01T00:01:00Z',
+      },
+      {
+        id: 'msg-4',
+        type: 'gemini',
+        content: [
+          { text: 'Understood. I will only save the durable preference.' },
+        ],
+        timestamp: '2026-01-01T00:01:05Z',
+      },
+    ],
+    prompt:
+      'Please save any persistent preferences or facts about me from our conversation to memory.',
+    assert: async (rig, result) => {
+      await rig.waitForToolCall('write_file').catch(() => {});
+      const writeCalls = rig
+        .readToolLogs()
+        .filter((log) =>
+          ['write_file', 'replace'].includes(log.toolRequest.name),
+        );
+
+      expect(
+        writeCalls.length,
+        'Expected memoryV2 save flow to edit a markdown memory file',
+      ).toBeGreaterThan(0);
+
+      await rig.run({
+        args: ['--list-sessions'],
+        approvalMode: 'yolo',
+        timeout: 120000,
+      });
+
+      const record = await waitForSessionScratchpad(
+        rig.homeDir!,
+        'memory-scratchpad-eval',
+      );
+      expect(
+        record?.memoryScratchpad,
+        'Expected the resumed session log to contain a memoryScratchpad after session summary generation',
+      ).toBeDefined();
+      expect(record?.memoryScratchpad?.version).toBe(1);
+      expect(
+        record?.memoryScratchpad?.toolSequence?.some((toolName) =>
+          ['write_file', 'replace'].includes(toolName),
+        ),
+        'Expected memoryScratchpad.toolSequence to include the markdown editing tool used for memory persistence',
+      ).toBe(true);
+      expect(
+        record?.memoryScratchpad?.touchedPaths?.length,
+        'Expected memoryScratchpad to capture at least one touched path',
+      ).toBeGreaterThan(0);
+      expect(
+        record?.memoryScratchpad?.workflowSummary,
+        'Expected memoryScratchpad.workflowSummary to be populated',
+      ).toMatch(/write_file|replace/i);
+
+      assertModelHasOutput(result);
+    },
+  });
+
   const memoryV2RoutesUserProject =
     'Agent routes personal-to-user project notes to user-project memory';
   evalTest('USUALLY_PASSES', {

evals/skill_extraction.eval.ts

@@ -6,21 +6,30 @@
 
 import fsp from 'node:fs/promises';
 import path from 'node:path';
-import { describe, expect } from 'vitest';
+import { describe, expect, it } from 'vitest';
 import {
   type Config,
   ApprovalMode,
+  type MemoryScratchpad,
   SESSION_FILE_PREFIX,
   getProjectHash,
   startMemoryService,
 } from '@google/gemini-cli-core';
-import { componentEvalTest } from './component-test-helper.js';
+import { ComponentRig, componentEvalTest } from './component-test-helper.js';
+import {
+  average,
+  averageNullable,
+  countMatchingIds,
+  roundStat,
+} from './statistics-helper.js';
+import { prepareWorkspace } from './test-helper.js';
 
 interface SeedSession {
   sessionId: string;
   summary: string;
   userTurns: string[];
   timestampOffsetMinutes: number;
+  memoryScratchpad?: MemoryScratchpad;
 }
 
 interface MessageRecord {
@@ -30,6 +39,81 @@ interface MessageRecord {
   content: Array<{ text: string }>;
 }
 
+interface SessionVersion {
+  sessionId: string;
+  lastUpdated: string;
+}
+
+interface ExtractionRunSnapshot {
+  sessionIds: string[];
+  skillsCreated: string[];
+  candidateSessions: SessionVersion[];
+  processedSessions: SessionVersion[];
+  turnCount?: number;
+  durationMs?: number;
+  terminateReason?: string;
+}
+
+interface ExtractionOutcome {
+  state: { runs: ExtractionRunSnapshot[] };
+  skillsDir: string;
+  skillBodies: string[];
+}
+
+interface SkillQualitySignal {
+  label: string;
+  pattern: RegExp;
+}
+
+interface ScratchpadRunMetrics {
+  turnCount: number | null;
+  durationMs: number | null;
+  terminateReason: string | null;
+  skillsCreated: number;
+  candidateSessions: number;
+  processedSessions: number;
+  relevantReads: number;
+  distractorReads: number;
+  totalReads: number;
+  recall: number;
+  precision: number;
+  signalScore: number;
+  skillQualityScore: number;
+  skillQualityMax: number;
+  skillQualityRatio: number;
+  missingQualitySignals: string[];
+}
+
+interface ScratchpadStatsTrial {
+  trial: number;
+  baseline: ScratchpadRunMetrics;
+  enhanced: ScratchpadRunMetrics;
+}
+
+interface ScratchpadStatsAggregate {
+  turnCountAvg: number | null;
+  durationMsAvg: number | null;
+  recallAvg: number;
+  precisionAvg: number;
+  signalScoreAvg: number;
+  relevantReadsAvg: number;
+  distractorReadsAvg: number;
+  skillsCreatedAvg: number;
+  skillQualityScoreAvg: number;
+  skillQualityRatioAvg: number;
+}
+
+interface ScratchpadStatsReport {
+  generatedAt: string;
+  trials: number;
+  aggregate: {
+    baseline: ScratchpadStatsAggregate;
+    enhanced: ScratchpadStatsAggregate;
+  };
+  deltas: ScratchpadStatsAggregate;
+  results: ScratchpadStatsTrial[];
+}
+
 const WORKSPACE_FILES = {
   'package.json': JSON.stringify(
     {
@@ -68,6 +152,143 @@ function buildMessages(userTurns: string[]): MessageRecord[] {
   ]);
 }
 
+function padTurns(turns: string[]): string[] {
+  if (turns.length >= 10) {
+    return turns;
+  }
+
+  const padded = [...turns];
+  for (let i = turns.length; i < 10; i++) {
+    padded.push(`${turns[i % turns.length]} (repeat ${i + 1})`);
+  }
+  return padded;
+}
+
+function createScratchpad(
+  workflowSummary: string,
+  touchedPaths: string[],
+  validationStatus: MemoryScratchpad['validationStatus'] = 'passed',
+): MemoryScratchpad {
+  return {
+    version: 1,
+    workflowSummary,
+    toolSequence: ['run_shell_command'],
+    touchedPaths,
+    validationStatus,
+  };
+}
+
+function createWorkflowComparisonSessions(withScratchpad: boolean): {
+  sessions: SeedSession[];
+  relevantSessionIds: string[];
+  distractorSessionIds: string[];
+} {
+  const relevantWorkflowSummary =
+    'run_shell_command -> run_shell_command | paths packages/cli/src/config/settings.ts, docs/settings.md | validated';
+
+  const relevantScratchpad = withScratchpad
+    ? createScratchpad(relevantWorkflowSummary, [
+        'packages/cli/src/config/settings.ts',
+        'docs/settings.md',
+      ])
+    : undefined;
+
+  const sessions: SeedSession[] = [
+    {
+      sessionId: 'hidden-settings-workflow-a',
+      summary: 'Prepare release notes for settings launch',
+      timestampOffsetMinutes: 420,
+      memoryScratchpad: relevantScratchpad,
+      userTurns: padTurns([
+        'When we add a new setting, the durable workflow is to regenerate the settings docs instead of editing them by hand.',
+        'The sequence that worked was npm run predocs:settings, npm run schema:settings, then npm run docs:settings.',
+        'Skipping predocs leaves stale defaults in the generated docs.',
+        'We verify the workflow by checking that both the schema output and docs update together.',
+        'This exact command order is the recurring workflow we use for settings changes.',
+      ]),
+    },
+    {
+      sessionId: 'hidden-settings-workflow-b',
+      summary: 'Investigate CI drift in generated config reference',
+      timestampOffsetMinutes: 390,
+      memoryScratchpad: relevantScratchpad,
+      userTurns: padTurns([
+        'The config reference drift was fixed by rerunning the standard settings regeneration workflow.',
+        'We again used npm run predocs:settings before npm run schema:settings and npm run docs:settings.',
+        'The recurring rule is never to hand-edit generated settings docs.',
+        'The validation step is to confirm the schema artifact and docs changed together after regeneration.',
+        'This is the same recurring workflow we use every time a setting changes.',
+      ]),
+    },
+    {
+      sessionId: 'distractor-release-notes',
+      summary: 'Prepare release notes for auth launch',
+      timestampOffsetMinutes: 360,
+      memoryScratchpad: undefined,
+      userTurns: padTurns([
+        'This release-notes task was one-off and just needed manual wording updates.',
+        'I edited CHANGELOG.md and docs/release-notes.md directly.',
+        'There was no reusable command sequence here beyond proofreading the copy.',
+        'This task should not become a standing workflow.',
+        'Once the wording landed, we were done.',
+      ]),
+    },
+    {
+      sessionId: 'distractor-ci-snapshots',
+      summary: 'Investigate CI drift in auth snapshots',
+      timestampOffsetMinutes: 330,
+      memoryScratchpad: undefined,
+      userTurns: padTurns([
+        'This auth snapshot issue was specific to a flaky test in CI.',
+        'The only commands we ran were npm test -- auth and an isolated snapshot update.',
+        'It was not the recurring settings-doc workflow.',
+        'Once the flaky snapshot passed, there was no broader reusable procedure.',
+        'Treat this as a one-off CI cleanup.',
+      ]),
+    },
+    {
+      sessionId: 'distractor-onboarding-docs',
+      summary: 'Refresh onboarding documentation copy',
+      timestampOffsetMinutes: 300,
+      memoryScratchpad: undefined,
+      userTurns: padTurns([
+        'This was just a docs wording cleanup in docs/onboarding.md.',
+        'No command sequence was involved.',
+        'We manually edited the copy and reviewed it.',
+        'There is no recurring operational workflow to capture here.',
+        'This should stay a one-off docs edit.',
+      ]),
+    },
+    {
+      sessionId: 'distractor-deploy-copy',
+      summary: 'Adjust deployment checklist wording',
+      timestampOffsetMinutes: 270,
+      memoryScratchpad: undefined,
+      userTurns: padTurns([
+        'This was a wording-only change to docs/deploy.md.',
+        'We did not run a reusable command sequence.',
+        'It should not become a skill.',
+        'The edit was only for this deploy checklist cleanup.',
+        'After the copy change, the task was complete.',
+      ]),
+    },
+  ];
+
+  return {
+    sessions,
+    relevantSessionIds: [
+      'hidden-settings-workflow-a',
+      'hidden-settings-workflow-b',
+    ],
+    distractorSessionIds: [
+      'distractor-release-notes',
+      'distractor-ci-snapshots',
+      'distractor-onboarding-docs',
+      'distractor-deploy-copy',
+    ],
+  };
+}
+
 async function seedSessions(
   config: Config,
   sessions: SeedSession[],
@@ -78,9 +299,10 @@ async function seedSessions(
   const projectRoot = config.storage.getProjectRoot();
 
   for (const session of sessions) {
-    const timestamp = new Date(
+    const sessionTimestamp = new Date(
       Date.now() - session.timestampOffsetMinutes * 60 * 1000,
-    )
+    );
+    const timestamp = sessionTimestamp
       .toISOString()
       .slice(0, 16)
       .replace(/:/g, '-');
@@ -89,8 +311,9 @@ async function seedSessions(
       sessionId: session.sessionId,
       projectHash: getProjectHash(projectRoot),
       summary: session.summary,
+      memoryScratchpad: session.memoryScratchpad,
       startTime: new Date(Date.now() - 7 * 60 * 60 * 1000).toISOString(),
-      lastUpdated: new Date(Date.now() - 4 * 60 * 60 * 1000).toISOString(),
+      lastUpdated: sessionTimestamp.toISOString(),
       messages: buildMessages(session.userTurns),
     };
 
@@ -101,10 +324,9 @@ async function seedSessions(
   }
 }
 
-async function runExtractionAndReadState(config: Config): Promise<{
-  state: { runs: Array<{ sessionIds: string[]; skillsCreated: string[] }> };
-  skillsDir: string;
-}> {
+async function runExtractionAndReadState(
+  config: Config,
+): Promise<ExtractionOutcome> {
   await startMemoryService(config);
 
   const memoryDir = config.storage.getProjectMemoryTempDir();
@@ -113,7 +335,15 @@ async function runExtractionAndReadState(config: Config): Promise<{
 
   const raw = await fsp.readFile(statePath, 'utf-8');
   const state = JSON.parse(raw) as {
-    runs?: Array<{ sessionIds?: string[]; skillsCreated?: string[] }>;
+    runs?: Array<{
+      sessionIds?: string[];
+      skillsCreated?: string[];
+      candidateSessions?: SessionVersion[];
+      processedSessions?: SessionVersion[];
+      turnCount?: number;
+      durationMs?: number;
+      terminateReason?: string;
+    }>;
   };
   if (!Array.isArray(state.runs) || state.runs.length === 0) {
     throw new Error('Skill extraction finished without writing any run state');
@@ -126,27 +356,292 @@ async function runExtractionAndReadState(config: Config): Promise<{
         skillsCreated: Array.isArray(run.skillsCreated)
           ? run.skillsCreated
           : [],
+        candidateSessions: Array.isArray(run.candidateSessions)
+          ? run.candidateSessions
+          : [],
+        processedSessions: Array.isArray(run.processedSessions)
+          ? run.processedSessions
+          : [],
+        turnCount:
+          typeof run.turnCount === 'number' ? run.turnCount : undefined,
+        durationMs:
+          typeof run.durationMs === 'number' ? run.durationMs : undefined,
+        terminateReason:
+          typeof run.terminateReason === 'string'
+            ? run.terminateReason
+            : undefined,
       })),
     },
     skillsDir,
+    skillBodies: await readSkillBodies(skillsDir),
   };
 }
 
+async function summarizeScratchpadRun(
+  outcome: ExtractionOutcome,
+  run: ExtractionRunSnapshot,
+  scenario: ReturnType<typeof createWorkflowComparisonSessions>,
+): Promise<ScratchpadRunMetrics> {
+  const relevantReads = countMatchingIds(
+    run.processedSessions,
+    scenario.relevantSessionIds,
+  );
+  const distractorReads = countMatchingIds(
+    run.processedSessions,
+    scenario.distractorSessionIds,
+  );
+  const totalReads = run.processedSessions.length;
+  const quality = scoreSkillQuality(
+    outcome.skillBodies,
+    SETTINGS_SKILL_QUALITY_SIGNALS,
+  );
+
+  return {
+    turnCount: run.turnCount ?? null,
+    durationMs: run.durationMs ?? null,
+    terminateReason: run.terminateReason ?? null,
+    skillsCreated: run.skillsCreated.length,
+    candidateSessions: run.candidateSessions.length,
+    processedSessions: totalReads,
+    relevantReads,
+    distractorReads,
+    totalReads,
+    recall: relevantReads / scenario.relevantSessionIds.length,
+    precision: totalReads === 0 ? 0 : relevantReads / totalReads,
+    signalScore: relevantReads - distractorReads,
+    skillQualityScore: quality.score,
+    skillQualityMax: quality.maxScore,
+    skillQualityRatio:
+      quality.maxScore === 0 ? 0 : quality.score / quality.maxScore,
+    missingQualitySignals: quality.missing,
+  };
+}
+
+function averageScratchpadRuns(
+  runs: ScratchpadRunMetrics[],
+): ScratchpadStatsAggregate {
+  return {
+    turnCountAvg: roundStat(averageNullable(runs.map((run) => run.turnCount))),
+    durationMsAvg: roundStat(
+      averageNullable(runs.map((run) => run.durationMs)),
+    ),
+    recallAvg: roundStat(average(runs.map((run) => run.recall))) ?? 0,
+    precisionAvg: roundStat(average(runs.map((run) => run.precision))) ?? 0,
+    signalScoreAvg: roundStat(average(runs.map((run) => run.signalScore))) ?? 0,
+    relevantReadsAvg:
+      roundStat(average(runs.map((run) => run.relevantReads))) ?? 0,
+    distractorReadsAvg:
+      roundStat(average(runs.map((run) => run.distractorReads))) ?? 0,
+    skillsCreatedAvg:
+      roundStat(average(runs.map((run) => run.skillsCreated))) ?? 0,
+    skillQualityScoreAvg:
+      roundStat(average(runs.map((run) => run.skillQualityScore))) ?? 0,
+    skillQualityRatioAvg:
+      roundStat(average(runs.map((run) => run.skillQualityRatio))) ?? 0,
+  };
+}
+
+function diffScratchpadAggregates(
+  baseline: ScratchpadStatsAggregate,
+  enhanced: ScratchpadStatsAggregate,
+): ScratchpadStatsAggregate {
+  return {
+    turnCountAvg:
+      baseline.turnCountAvg === null || enhanced.turnCountAvg === null
+        ? null
+        : roundStat(enhanced.turnCountAvg - baseline.turnCountAvg),
+    durationMsAvg:
+      baseline.durationMsAvg === null || enhanced.durationMsAvg === null
+        ? null
+        : roundStat(enhanced.durationMsAvg - baseline.durationMsAvg),
+    recallAvg: roundStat(enhanced.recallAvg - baseline.recallAvg) ?? 0,
+    precisionAvg: roundStat(enhanced.precisionAvg - baseline.precisionAvg) ?? 0,
+    signalScoreAvg:
+      roundStat(enhanced.signalScoreAvg - baseline.signalScoreAvg) ?? 0,
+    relevantReadsAvg:
+      roundStat(enhanced.relevantReadsAvg - baseline.relevantReadsAvg) ?? 0,
+    distractorReadsAvg:
+      roundStat(enhanced.distractorReadsAvg - baseline.distractorReadsAvg) ?? 0,
+    skillsCreatedAvg:
+      roundStat(enhanced.skillsCreatedAvg - baseline.skillsCreatedAvg) ?? 0,
+    skillQualityScoreAvg:
+      roundStat(
+        enhanced.skillQualityScoreAvg - baseline.skillQualityScoreAvg,
+      ) ?? 0,
+    skillQualityRatioAvg:
+      roundStat(
+        enhanced.skillQualityRatioAvg - baseline.skillQualityRatioAvg,
+      ) ?? 0,
+  };
+}
+
+async function runScenarioWithFreshRig(
+  sessions: SeedSession[],
+): Promise<ExtractionOutcome> {
+  const rig = new ComponentRig({
+    configOverrides: EXTRACTION_CONFIG_OVERRIDES,
+  });
+  try {
+    await rig.initialize();
+    await prepareWorkspace(rig.testDir, rig.testDir, WORKSPACE_FILES);
+    await seedSessions(rig.config!, sessions);
+    return await runExtractionAndReadState(rig.config!);
+  } finally {
+    await rig.cleanup();
+  }
+}
+
+async function runScratchpadStatsTrial(
+  trial: number,
+): Promise<ScratchpadStatsTrial> {
+  const baselineScenario = createWorkflowComparisonSessions(false);
+  const enhancedScenario = createWorkflowComparisonSessions(true);
+
+  const baselineOutcome = await runScenarioWithFreshRig(
+    baselineScenario.sessions,
+  );
+  const enhancedOutcome = await runScenarioWithFreshRig(
+    enhancedScenario.sessions,
+  );
+
+  const baselineRun = baselineOutcome.state.runs.at(-1);
+  const enhancedRun = enhancedOutcome.state.runs.at(-1);
+  if (!baselineRun || !enhancedRun) {
+    throw new Error('Expected both baseline and scratchpad runs to exist');
+  }
+
+  expectSuccessfulExtractionRun(baselineRun);
+  expectSuccessfulExtractionRun(enhancedRun);
+
+  return {
+    trial,
+    baseline: await summarizeScratchpadRun(
+      baselineOutcome,
+      baselineRun,
+      baselineScenario,
+    ),
+    enhanced: await summarizeScratchpadRun(
+      enhancedOutcome,
+      enhancedRun,
+      enhancedScenario,
+    ),
+  };
+}
+
+async function runScratchpadStatsReport(
+  trials: number,
+): Promise<ScratchpadStatsReport> {
+  const results: ScratchpadStatsTrial[] = [];
+
+  for (let trial = 1; trial <= trials; trial++) {
+    results.push(await runScratchpadStatsTrial(trial));
+  }
+
+  const baseline = averageScratchpadRuns(
+    results.map((result) => result.baseline),
+  );
+  const enhanced = averageScratchpadRuns(
+    results.map((result) => result.enhanced),
+  );
+
+  return {
+    generatedAt: new Date().toISOString(),
+    trials,
+    aggregate: {
+      baseline,
+      enhanced,
+    },
+    deltas: diffScratchpadAggregates(baseline, enhanced),
+    results,
+  };
+}
+
+async function writeScratchpadStatsReport(
+  report: ScratchpadStatsReport,
+): Promise<string> {
+  const outputPath = path.resolve(
+    process.cwd(),
+    'evals/logs/skill_extraction_scratchpad_stats.json',
+  );
+  await fsp.mkdir(path.dirname(outputPath), { recursive: true });
+  await fsp.writeFile(outputPath, `${JSON.stringify(report, null, 2)}\n`);
+  return outputPath;
+}
+
 async function readSkillBodies(skillsDir: string): Promise<string[]> {
+  const bodies: string[] = [];
+
   try {
     const entries = await fsp.readdir(skillsDir, { withFileTypes: true });
-    const skillDirs = entries.filter((entry) => entry.isDirectory());
-    const bodies = await Promise.all(
-      skillDirs.map((entry) =>
-        fsp.readFile(path.join(skillsDir, entry.name, 'SKILL.md'), 'utf-8'),
-      ),
-    );
+    for (const entry of entries) {
+      if (!entry.isDirectory()) {
+        continue;
+      }
+
+      try {
+        bodies.push(
+          await fsp.readFile(
+            path.join(skillsDir, entry.name, 'SKILL.md'),
+            'utf-8',
+          ),
+        );
+      } catch {
+        // Ignore incomplete skill directories so one bad artifact does not hide
+        // valid skills created in the same eval run.
+      }
+    }
     return bodies;
   } catch {
     return [];
   }
 }
 
+function expectSuccessfulExtractionRun(run: ExtractionRunSnapshot): void {
+  expect(run.turnCount).toBeGreaterThan(0);
+  expect(run.turnCount).toBeLessThanOrEqual(30);
+  expect(run.durationMs).toBeGreaterThan(0);
+  expect(run.terminateReason).toBe('GOAL');
+}
+
+function scoreSkillQuality(
+  skillBodies: string[],
+  signals: SkillQualitySignal[],
+): { score: number; maxScore: number; missing: string[] } {
+  const combined = skillBodies.join('\n\n');
+  const matched = signals.filter((signal) => signal.pattern.test(combined));
+
+  return {
+    score: matched.length,
+    maxScore: signals.length,
+    missing: signals
+      .filter((signal) => !signal.pattern.test(combined))
+      .map((signal) => signal.label),
+  };
+}
+
+const SETTINGS_SKILL_QUALITY_SIGNALS: SkillQualitySignal[] = [
+  { label: 'predocs command', pattern: /npm run predocs:settings/i },
+  { label: 'schema command', pattern: /npm run schema:settings/i },
+  { label: 'docs command', pattern: /npm run docs:settings/i },
+  { label: 'verification guidance', pattern: /verif(?:y|ication)/i },
+  {
+    label: 'generated docs warning or ordering constraint',
+    pattern:
+      /do not hand-edit|manual edits|exact command order|preserve.*order/i,
+  },
+];
+
+const DB_MIGRATION_SKILL_QUALITY_SIGNALS: SkillQualitySignal[] = [
+  { label: 'db check command', pattern: /npm run db:check/i },
+  { label: 'db migrate command', pattern: /npm run db:migrate/i },
+  { label: 'db validate command', pattern: /npm run db:validate/i },
+  { label: 'rollback guidance', pattern: /npm run db:rollback|rollback/i },
+  {
+    label: 'ordering constraint',
+    pattern: /check.*migrate.*validate|ordering is critical|mandatory/i,
+  },
+];
+
 /**
  * Shared configOverrides for all skill extraction component evals.
  * - experimentalAutoMemory: enables the Auto Memory skill extraction pipeline.
@@ -158,6 +653,16 @@ const EXTRACTION_CONFIG_OVERRIDES = {
   approvalMode: ApprovalMode.YOLO,
 };
 
+function parseScratchpadStatsTrials(): number {
+  const configured = Number.parseInt(
+    process.env['SCRATCHPAD_STATS_TRIALS'] ?? '8',
+    10,
+  );
+  return Number.isFinite(configured) && configured > 0 ? configured : 8;
+}
+
+const SCRATCHPAD_STATS_TRIALS = parseScratchpadStatsTrials();
+
 describe('Skill Extraction', () => {
   componentEvalTest('USUALLY_PASSES', {
     suiteName: 'skill-extraction',
@@ -264,15 +769,24 @@ describe('Skill Extraction', () => {
       const { state, skillsDir } = await runExtractionAndReadState(config);
       const skillBodies = await readSkillBodies(skillsDir);
       const combinedSkills = skillBodies.join('\n\n');
+      const quality = scoreSkillQuality(
+        skillBodies,
+        SETTINGS_SKILL_QUALITY_SIGNALS,
+      );
 
       expect(state.runs).toHaveLength(1);
       expect(state.runs[0].sessionIds).toHaveLength(2);
+      expectSuccessfulExtractionRun(state.runs[0]);
       expect(state.runs[0].skillsCreated.length).toBeGreaterThanOrEqual(1);
       expect(skillBodies.length).toBeGreaterThanOrEqual(1);
       expect(combinedSkills).toContain('npm run predocs:settings');
       expect(combinedSkills).toContain('npm run schema:settings');
       expect(combinedSkills).toContain('npm run docs:settings');
-      expect(combinedSkills).toMatch(/Verification/i);
+      expect(combinedSkills).toMatch(/verif(?:y|ication)/i);
+      expect(
+        quality.score,
+        `missing quality signals: ${quality.missing.join(', ')}`,
+      ).toBeGreaterThanOrEqual(4);
 
       // Verify the extraction agent activated skill-creator for design guidance.
       expect(config.getSkillManager().isSkillActive('skill-creator')).toBe(
@@ -281,6 +795,96 @@ describe('Skill Extraction', () => {
     },
   });
 
+  componentEvalTest('USUALLY_PASSES', {
+    suiteName: 'skill-extraction',
+    suiteType: 'component-level',
+    name: 'memory scratchpad improves repeated-workflow recall versus summary-only index',
+    files: WORKSPACE_FILES,
+    timeout: 360000,
+    configOverrides: EXTRACTION_CONFIG_OVERRIDES,
+    assert: async () => {
+      const baselineScenario = createWorkflowComparisonSessions(false);
+      const enhancedScenario = createWorkflowComparisonSessions(true);
+
+      const baselineOutcome = await runScenarioWithFreshRig(
+        baselineScenario.sessions,
+      );
+      const enhancedOutcome = await runScenarioWithFreshRig(
+        enhancedScenario.sessions,
+      );
+
+      const baselineRun = baselineOutcome.state.runs.at(-1);
+      const enhancedRun = enhancedOutcome.state.runs.at(-1);
+      if (!baselineRun || !enhancedRun) {
+        throw new Error('Expected both baseline and scratchpad runs to exist');
+      }
+
+      expectSuccessfulExtractionRun(baselineRun);
+      expectSuccessfulExtractionRun(enhancedRun);
+
+      const baselineRelevantReads = countMatchingIds(
+        baselineRun.processedSessions,
+        baselineScenario.relevantSessionIds,
+      );
+      const enhancedRelevantReads = countMatchingIds(
+        enhancedRun.processedSessions,
+        enhancedScenario.relevantSessionIds,
+      );
+      const baselineDistractorReads = countMatchingIds(
+        baselineRun.processedSessions,
+        baselineScenario.distractorSessionIds,
+      );
+      const enhancedDistractorReads = countMatchingIds(
+        enhancedRun.processedSessions,
+        enhancedScenario.distractorSessionIds,
+      );
+      const baselineSignalScore =
+        baselineRelevantReads - baselineDistractorReads;
+      const enhancedSignalScore =
+        enhancedRelevantReads - enhancedDistractorReads;
+
+      expect(enhancedRun.candidateSessions).toHaveLength(
+        enhancedScenario.sessions.length,
+      );
+      expect(enhancedRelevantReads).toBeGreaterThanOrEqual(2);
+      expect(enhancedRelevantReads).toBeGreaterThanOrEqual(
+        baselineRelevantReads,
+      );
+      expect(enhancedDistractorReads).toBeLessThanOrEqual(
+        baselineDistractorReads,
+      );
+      expect(enhancedSignalScore).toBeGreaterThan(baselineSignalScore);
+    },
+  });
+
+  if (process.env['RUN_SCRATCHPAD_STATS'] === '1') {
+    componentEvalTest('USUALLY_PASSES', {
+      suiteName: 'skill-extraction',
+      suiteType: 'component-level',
+      name: 'reports memory scratchpad retrieval statistics',
+      timeout: Math.max(360000, SCRATCHPAD_STATS_TRIALS * 150000),
+      configOverrides: EXTRACTION_CONFIG_OVERRIDES,
+      assert: async () => {
+        const report = await runScratchpadStatsReport(SCRATCHPAD_STATS_TRIALS);
+        const outputPath = await writeScratchpadStatsReport(report);
+
+        console.info(
+          `Wrote scratchpad stats report to ${outputPath}\n${JSON.stringify(
+            report.aggregate,
+            null,
+            2,
+          )}`,
+        );
+
+        expect(report.results).toHaveLength(SCRATCHPAD_STATS_TRIALS);
+        expect(report.aggregate.baseline.recallAvg).toBeGreaterThan(0);
+        expect(report.aggregate.enhanced.recallAvg).toBeGreaterThan(0);
+      },
+    });
+  } else {
+    it.skip('reports memory scratchpad retrieval statistics', () => {});
+  }
+
   componentEvalTest('USUALLY_PASSES', {
     suiteName: 'skill-extraction',
     suiteType: 'component-level',
@@ -330,15 +934,24 @@ describe('Skill Extraction', () => {
       const { state, skillsDir } = await runExtractionAndReadState(config);
       const skillBodies = await readSkillBodies(skillsDir);
       const combinedSkills = skillBodies.join('\n\n');
+      const quality = scoreSkillQuality(
+        skillBodies,
+        DB_MIGRATION_SKILL_QUALITY_SIGNALS,
+      );
 
       expect(state.runs).toHaveLength(1);
       expect(state.runs[0].sessionIds).toHaveLength(2);
+      expectSuccessfulExtractionRun(state.runs[0]);
       expect(state.runs[0].skillsCreated.length).toBeGreaterThanOrEqual(1);
       expect(skillBodies.length).toBeGreaterThanOrEqual(1);
       expect(combinedSkills).toContain('npm run db:check');
       expect(combinedSkills).toContain('npm run db:migrate');
       expect(combinedSkills).toContain('npm run db:validate');
       expect(combinedSkills).toMatch(/rollback/i);
+      expect(
+        quality.score,
+        `missing quality signals: ${quality.missing.join(', ')}`,
+      ).toBeGreaterThanOrEqual(4);
 
       // Verify the extraction agent activated skill-creator for design guidance.
       expect(config.getSkillManager().isSkillActive('skill-creator')).toBe(

evals/statistics-helper.ts

@@ -0,0 +1,26 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+export function countMatchingIds<T extends { sessionId: string }>(
+  items: T[],
+  expectedIds: string[],
+): number {
+  const expected = new Set(expectedIds);
+  return items.filter((item) => expected.has(item.sessionId)).length;
+}
+
+export function roundStat(value: number | null): number | null {
+  return value === null ? null : Number(value.toFixed(4));
+}
+
+export function average(values: number[]): number {
+  return values.reduce((sum, value) => sum + value, 0) / values.length;
+}
+
+export function averageNullable(values: Array<number | null>): number | null {
+  const numericValues = values.filter((value) => value !== null);
+  return numericValues.length === 0 ? null : average(numericValues);
+}

evals/tracker.eval.ts

@@ -62,11 +62,13 @@ describe('tracker_mode', () => {
         'Expected tracker_update_task tool to be called',
       ).toBe(true);
 
-      const updateCall = toolLogs.find(
+      const updateCalls = toolLogs.filter(
         (log) => log.toolRequest.name === TRACKER_UPDATE_TASK_TOOL_NAME,
       );
-      expect(updateCall).toBeDefined();
-      const updateArgs = JSON.parse(updateCall!.toolRequest.args);
+      expect(updateCalls.length).toBeGreaterThan(0);
+      const updateArgs = JSON.parse(
+        updateCalls[updateCalls.length - 1].toolRequest.args,
+      );
       expect(updateArgs.status).toBe('closed');
 
       const loginContent = fs.readFileSync(
@@ -128,12 +130,52 @@ describe('tracker_mode', () => {
     prompt:
       'Where is my task tracker storage located? Please provide the absolute path in your response.',
     assert: async (rig, result) => {
-      // The rig sets GEMINI_CLI_HOME to rig.homeDir
-      const homeDir = rig.homeDir!;
-      // The response should contain the dynamic path which includes the home directory
-      // and follows the .gemini/tmp/.../tracker structure.
-      expect(result).toContain(homeDir);
+      // The response should contain the dynamic path which follows the .gemini/tmp/.../tracker structure.
       expect(result).toMatch(/\.gemini\/tmp\/.*\/tracker/);
     },
   });
+
+  evalTest('USUALLY_PASSES', {
+    suiteName: 'default',
+    suiteType: 'behavioral',
+    name: 'should update the tracker in the same turn as the task completion to save turns',
+    params: {
+      settings: { experimental: { taskTracker: true } },
+    },
+    files: FILES,
+    prompt:
+      'We have a bug in src/login.js: the password check is missing. Fix this bug. Then, create a new file src/auth.js that exports a simple verifyToken function. Please organize this into tasks and execute them.',
+    assert: async (rig, result) => {
+      await rig.waitForToolCall(TRACKER_CREATE_TASK_TOOL_NAME);
+      await rig.waitForToolCall(TRACKER_UPDATE_TASK_TOOL_NAME);
+
+      const toolLogs = rig.readToolLogs();
+
+      // Get the prompt ID of the fix for login.js
+      const loginEditCalls = toolLogs.filter(
+        (log) =>
+          (log.toolRequest.name === 'replace' ||
+            log.toolRequest.name === 'write_file') &&
+          log.toolRequest.args.includes('login.js'),
+      );
+
+      expect(loginEditCalls.length).toBeGreaterThan(0);
+      const loginEditPromptId =
+        loginEditCalls[loginEditCalls.length - 1].toolRequest.prompt_id;
+
+      // Verify there is an update to the tracker in the exact same turn
+      const parallelTrackerUpdates = toolLogs.filter(
+        (log) =>
+          log.toolRequest.name === TRACKER_UPDATE_TASK_TOOL_NAME &&
+          log.toolRequest.prompt_id === loginEditPromptId,
+      );
+
+      expect(
+        parallelTrackerUpdates.length,
+        'Expected tracker_update_task to be called in the same turn as the login.js fix',
+      ).toBeGreaterThan(0);
+
+      assertModelHasOutput(result);
+    },
+  });
 });

integration-tests/voice-mode.test.ts

@@ -0,0 +1,76 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import { describe, it, expect, beforeEach, afterEach } from 'vitest';
+import { TestRig } from './test-helper.js';
+import {
+  WhisperModelManager,
+  WhisperTranscriptionProvider,
+} from '@google/gemini-cli-core';
+import * as fs from 'node:fs';
+import commandExists from 'command-exists';
+
+describe('Voice Mode Integration', () => {
+  let rig: TestRig;
+
+  beforeEach(() => {
+    rig = new TestRig();
+  });
+
+  afterEach(async () => await rig.cleanup());
+
+  it('should be able to download tiny whisper model', async () => {
+    // This test doesn't require the binary, only network access.
+    // However, it's slow and downloads 75MB. We'll keep it for now but
+    // wrap it in a try-catch to avoid failing on network flakiness in CI.
+    const manager = new WhisperModelManager();
+    const modelName = 'ggml-tiny.en.bin';
+
+    try {
+      // Cleanup if already exists to ensure we actually test download
+      const modelPath = manager.getModelPath(modelName);
+      if (fs.existsSync(modelPath)) {
+        fs.unlinkSync(modelPath);
+      }
+
+      await manager.downloadModel(modelName);
+      expect(fs.existsSync(modelPath)).toBe(true);
+      expect(fs.statSync(modelPath).size).toBeGreaterThan(70 * 1024 * 1024); // ~75MB
+    } catch (e) {
+      console.warn(
+        'Skipping whisper model download test due to error (possibly network):',
+        e,
+      );
+    }
+  }, 300000); // 5 min timeout for download
+
+  it('should initialize WhisperTranscriptionProvider and handle process', async () => {
+    // Skip this test if whisper-stream is not installed (typical for CI)
+    try {
+      await commandExists('whisper-stream');
+    } catch {
+      console.log(
+        'Skipping Whisper transcription test: whisper-stream not found',
+      );
+      return;
+    }
+
+    const manager = new WhisperModelManager();
+    const modelName = 'ggml-tiny.en.bin';
+    if (!manager.isModelInstalled(modelName)) {
+      await manager.downloadModel(modelName);
+    }
+
+    const provider = new WhisperTranscriptionProvider({
+      modelPath: manager.getModelPath(modelName),
+    });
+
+    // Since we can't easily provide real mic input in CI,
+    // we just verify it can start and be disconnected.
+    await provider.connect();
+    provider.disconnect();
+  });
+});

package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "@google/gemini-cli",
-  "version": "0.41.0-nightly.20260423.gaa05b4583",
+  "version": "0.42.0-nightly.20260428.g59b2dea0e",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "@google/gemini-cli",
-      "version": "0.41.0-nightly.20260423.gaa05b4583",
+      "version": "0.42.0-nightly.20260428.g59b2dea0e",
       "workspaces": [
         "packages/*"
       ],
@@ -18077,7 +18077,7 @@
     },
     "packages/a2a-server": {
       "name": "@google/gemini-cli-a2a-server",
-      "version": "0.41.0-nightly.20260423.gaa05b4583",
+      "version": "0.42.0-nightly.20260428.g59b2dea0e",
       "dependencies": {
         "@a2a-js/sdk": "0.3.11",
         "@google-cloud/storage": "^7.16.0",
@@ -18206,7 +18206,7 @@
     },
     "packages/cli": {
       "name": "@google/gemini-cli",
-      "version": "0.41.0-nightly.20260423.gaa05b4583",
+      "version": "0.42.0-nightly.20260428.g59b2dea0e",
       "license": "Apache-2.0",
       "dependencies": {
         "@agentclientprotocol/sdk": "^0.16.1",
@@ -18354,7 +18354,7 @@
     },
     "packages/core": {
       "name": "@google/gemini-cli-core",
-      "version": "0.41.0-nightly.20260423.gaa05b4583",
+      "version": "0.42.0-nightly.20260428.g59b2dea0e",
       "license": "Apache-2.0",
       "dependencies": {
         "@a2a-js/sdk": "0.3.11",
@@ -18390,6 +18390,7 @@
         "ajv-formats": "^3.0.0",
         "chardet": "^2.1.0",
         "chokidar": "^5.0.0",
+        "command-exists": "^1.2.9",
         "diff": "^8.0.3",
         "dotenv": "^17.2.4",
         "dotenv-expand": "^12.0.3",
@@ -18664,7 +18665,7 @@
     },
     "packages/devtools": {
       "name": "@google/gemini-cli-devtools",
-      "version": "0.41.0-nightly.20260423.gaa05b4583",
+      "version": "0.42.0-nightly.20260428.g59b2dea0e",
       "license": "Apache-2.0",
       "dependencies": {
         "ws": "^8.16.0"
@@ -18679,7 +18680,7 @@
     },
     "packages/sdk": {
       "name": "@google/gemini-cli-sdk",
-      "version": "0.41.0-nightly.20260423.gaa05b4583",
+      "version": "0.42.0-nightly.20260428.g59b2dea0e",
       "license": "Apache-2.0",
       "dependencies": {
         "@google/gemini-cli-core": "file:../core",
@@ -18710,7 +18711,7 @@
     },
     "packages/test-utils": {
       "name": "@google/gemini-cli-test-utils",
-      "version": "0.41.0-nightly.20260423.gaa05b4583",
+      "version": "0.42.0-nightly.20260428.g59b2dea0e",
       "license": "Apache-2.0",
       "dependencies": {
         "@google/gemini-cli-core": "file:../core",
@@ -18742,7 +18743,7 @@
     },
     "packages/vscode-ide-companion": {
       "name": "gemini-cli-vscode-ide-companion",
-      "version": "0.41.0-nightly.20260423.gaa05b4583",
+      "version": "0.42.0-nightly.20260428.g59b2dea0e",
       "license": "LICENSE",
       "dependencies": {
         "@modelcontextprotocol/sdk": "^1.23.0",

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@google/gemini-cli",
-  "version": "0.41.0-nightly.20260423.gaa05b4583",
+  "version": "0.42.0-nightly.20260428.g59b2dea0e",
   "engines": {
     "node": ">=20.0.0"
   },
@@ -14,7 +14,7 @@
     "url": "git+https://github.com/google-gemini/gemini-cli.git"
   },
   "config": {
-    "sandboxImageUri": "us-docker.pkg.dev/gemini-code-dev/gemini-cli/sandbox:0.41.0-nightly.20260423.gaa05b4583"
+    "sandboxImageUri": "us-docker.pkg.dev/gemini-code-dev/gemini-cli/sandbox:0.42.0-nightly.20260428.g59b2dea0e"
   },
   "scripts": {
     "start": "cross-env NODE_ENV=development node scripts/start.js",
@@ -63,7 +63,6 @@
     "lint:all": "node scripts/lint.js",
     "format": "prettier --experimental-cli --write .",
     "typecheck": "npm run typecheck --workspaces --if-present && tsc -b evals/tsconfig.json integration-tests/tsconfig.json memory-tests/tsconfig.json",
-    "metrics": "tsx tools/gemini-cli-bot/metrics/index.ts",
     "preflight": "npm run clean && npm ci && npm run format && npm run build && npm run lint:ci && npm run typecheck && npm run test:ci",
     "prepare": "husky && npm run bundle",
     "prepare:package": "node scripts/prepare-package.js",

packages/a2a-server/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@google/gemini-cli-a2a-server",
-  "version": "0.41.0-nightly.20260423.gaa05b4583",
+  "version": "0.42.0-nightly.20260428.g59b2dea0e",
   "description": "Gemini CLI A2A Server",
   "repository": {
     "type": "git",

packages/cli/index.ts

@@ -9,6 +9,11 @@
 import { spawn } from 'node:child_process';
 import os from 'node:os';
 import v8 from 'node:v8';
+import {
+  RELAUNCH_EXIT_CODE,
+  getSpawnConfig,
+  getScriptArgs,
+} from './src/utils/processUtils.js';
 
 // --- Global Entry Point ---
 
@@ -74,18 +79,10 @@ async function run() {
     // --- Lightweight Parent Process / Daemon ---
     // We avoid importing heavy dependencies here to save ~1.5s of startup time.
 
-    const nodeArgs: string[] = [...process.execArgv];
-    const scriptArgs = process.argv.slice(2);
-
+    const scriptArgs = getScriptArgs();
     const memoryArgs = await getMemoryNodeArgs();
-    nodeArgs.push(...memoryArgs);
+    const { spawnArgs, env: newEnv } = getSpawnConfig(memoryArgs, scriptArgs);
 
-    const script = process.argv[1];
-    nodeArgs.push(script);
-    nodeArgs.push(...scriptArgs);
-
-    const newEnv = { ...process.env, GEMINI_CLI_NO_RELAUNCH: 'true' };
-    const RELAUNCH_EXIT_CODE = 199;
     let latestAdminSettings: unknown = undefined;
 
     // Prevent the parent process from exiting prematurely on signals.
@@ -97,7 +94,7 @@ async function run() {
     const runner = () => {
       process.stdin.pause();
 
-      const child = spawn(process.execPath, nodeArgs, {
+      const child = spawn(process.execPath, spawnArgs, {
         stdio: ['inherit', 'inherit', 'inherit', 'ipc'],
         env: newEnv,
       });

packages/cli/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@google/gemini-cli",
-  "version": "0.41.0-nightly.20260423.gaa05b4583",
+  "version": "0.42.0-nightly.20260428.g59b2dea0e",
   "description": "Gemini CLI",
   "license": "Apache-2.0",
   "repository": {
@@ -27,7 +27,7 @@
     "dist"
   ],
   "config": {
-    "sandboxImageUri": "us-docker.pkg.dev/gemini-code-dev/gemini-cli/sandbox:0.41.0-nightly.20260423.gaa05b4583"
+    "sandboxImageUri": "us-docker.pkg.dev/gemini-code-dev/gemini-cli/sandbox:0.42.0-nightly.20260428.g59b2dea0e"
   },
   "dependencies": {
     "@agentclientprotocol/sdk": "^0.16.1",

packages/cli/src/acp/README.md

@@ -0,0 +1,81 @@
+# Agent Client Protocol (ACP) Implementation
+
+This directory contains the implementation of the Agent Client Protocol (ACP)
+for the Gemini CLI. The ACP allows external clients (like IDE extensions) to
+communicate with the Gemini CLI agent over a structured JSON-RPC based protocol.
+
+## Directory Structure
+
+Following Phase 1 of the modularization refactor, the ACP client is organized
+into the following specialized modules, all sharing the `acp` prefix for
+consistency:
+
+- **[acpStdioTransport.ts](./acpStdioTransport.ts)**: Handles raw I/O. It sets
+  up the Web streams for standard input/output and creates the
+  `AgentSideConnection` using line-delimited JSON (ndjson).
+- **[acpRpcDispatcher.ts](./acpRpcDispatcher.ts)**: Contains the `GeminiAgent`
+  class. This is the main entry point for incoming JSON-RPC messages. It
+  implements the protocol methods and delegates session-specific work to the
+  manager and individual sessions.
+- **[acpSessionManager.ts](./acpSessionManager.ts)**: Manages multi-session
+  state. It handles session creation (`newSession`), loading (`loadSession`),
+  and configuration, isolating session state from the RPC routing.
+- **[acpSession.ts](./acpSession.ts)**: Manages individual active chat sessions.
+  It handles prompt execution, `@path` file resolution, tool execution, command
+  interception, and streaming updates back to the client.
+- **[acpUtils.ts](./acpUtils.ts)**: Contains shared helper functions, type
+  mappers (e.g., mapping internal tool kinds to ACP kinds), and Zod schemas used
+  across the modules.
+- **[acpErrors.ts](./acpErrors.ts)**: Centralized error handling and mapping to
+  ACP-compliant error codes.
+- **[acpCommandHandler.ts](./acpCommandHandler.ts)**: Handles interception and
+  execution of slash commands (e.g., `/memory`, `/init`) sent via ACP prompts.
+- **[acpFileSystemService.ts](./acpFileSystemService.ts)**: Provides access to
+  the file system restricted by the workspace boundaries and permissions.
+
+## Development Instructions
+
+### Running Tests
+
+Tests are co-located with the source files:
+
+- `acpRpcDispatcher.test.ts`: Tests for initialization, authentication, and
+  handler delegation.
+- `acpSessionManager.test.ts`: Tests for session lifecycle and configuration.
+- `acpSession.test.ts`: Tests for prompt loops, tool execution, and @path
+  resolution.
+- `acpResume.test.ts`: Integration tests for loading/resuming sessions.
+
+To run specific tests, use Vitest with the workspace filter:
+
+```bash
+# General pattern
+npm test -w @google/gemini-cli -- src/acp/<test-file-name>.ts
+
+# Example
+npm test -w @google/gemini-cli -- src/acp/acpRpcDispatcher.test.ts
+```
+
+Note: You may need to ensure your environment has Node available. If running in
+a restricted environment, try sourcing NVM first:
+
+```bash
+source ~/.nvm/nvm.sh && nvm use default && npm test -w @google/gemini-cli -- src/acp/acpSession.test.ts
+```
+
+### Adding New Features
+
+- **New RPC Method**: Add the method to `GeminiAgent` in `acpRpcDispatcher.ts`
+  and register it in the `AgentSideConnection` setup if necessary.
+- **Session State**: If a feature requires storing state across turns within a
+  session, add it to the `Session` class in `acpSession.ts`.
+- **Protocol Helpers**: Add any new mapping or serialization logic to
+  `acpUtils.ts`.
+
+### Coding Conventions
+
+- **Imports**: Use specific imports and do not import across package boundaries
+  using relative paths.
+- **License Headers**: All new files must include the Apache-2.0 license header.
+- **Type Safety**: Avoid using `any` assertions. Use Zod schemas to validate
+  untrusted input from the protocol.

packages/cli/src/acp/acpCommandHandler.test.ts

@@ -4,7 +4,7 @@
  * SPDX-License-Identifier: Apache-2.0
  */
 
-import { CommandHandler } from './commandHandler.js';
+import { CommandHandler } from './acpCommandHandler.js';
 import { describe, it, expect } from 'vitest';
 
 describe('CommandHandler', () => {

packages/cli/src/acp/acpCommandHandler.ts

@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2025 Google LLC
+ * Copyright 2026 Google LLC
  * SPDX-License-Identifier: Apache-2.0
  */
 

packages/cli/src/acp/acpErrors.test.ts

@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2025 Google LLC
+ * Copyright 2026 Google LLC
  * SPDX-License-Identifier: Apache-2.0
  */
 

packages/cli/src/acp/acpErrors.ts

@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2025 Google LLC
+ * Copyright 2026 Google LLC
  * SPDX-License-Identifier: Apache-2.0
  */
 

packages/cli/src/acp/acpFileSystemService.test.ts

@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2025 Google LLC
+ * Copyright 2026 Google LLC
  * SPDX-License-Identifier: Apache-2.0
  */
 
@@ -13,7 +13,7 @@ import {
   afterEach,
   type Mocked,
 } from 'vitest';
-import { AcpFileSystemService } from './fileSystemService.js';
+import { AcpFileSystemService } from './acpFileSystemService.js';
 import type { AgentSideConnection } from '@agentclientprotocol/sdk';
 import type { FileSystemService } from '@google/gemini-cli-core';
 import os from 'node:os';

packages/cli/src/acp/acpFileSystemService.ts

@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2025 Google LLC
+ * Copyright 2026 Google LLC
  * SPDX-License-Identifier: Apache-2.0
  */
 

packages/cli/src/acp/acpResume.test.ts

@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2025 Google LLC
+ * Copyright 2026 Google LLC
  * SPDX-License-Identifier: Apache-2.0
  */
 
@@ -13,7 +13,7 @@ import {
   type Mocked,
   type Mock,
 } from 'vitest';
-import { GeminiAgent } from './acpClient.js';
+import { GeminiAgent } from './acpRpcDispatcher.js';
 import * as acp from '@agentclientprotocol/sdk';
 import {
   ApprovalMode,
@@ -28,6 +28,7 @@ import {
 } from '../utils/sessionUtils.js';
 import { convertSessionToClientHistory } from '@google/gemini-cli-core';
 import type { LoadedSettings } from '../config/settings.js';
+import { waitFor } from '../test-utils/async.js';
 
 vi.mock('../config/config.js', () => ({
   loadCliConfig: vi.fn(),
@@ -106,6 +107,9 @@ describe('GeminiAgent Session Resume', () => {
       getHasAccessToPreviewModel: vi.fn().mockReturnValue(false),
       getGemini31LaunchedSync: vi.fn().mockReturnValue(false),
       getCheckpointingEnabled: vi.fn().mockReturnValue(false),
+      toolRegistry: {
+        getTool: vi.fn().mockReturnValue({ kind: 'read' }),
+      },
       get config() {
         return this;
       },
@@ -170,11 +174,6 @@ describe('GeminiAgent Session Resume', () => {
       ],
     };
 
-    // eslint-disable-next-line @typescript-eslint/no-explicit-any
-    (mockConfig as any).toolRegistry = {
-      getTool: vi.fn().mockReturnValue({ kind: 'read' }),
-    };
-
     (SessionSelector as unknown as Mock).mockImplementation(() => ({
       resolveSession: vi.fn().mockResolvedValue({
         sessionData,
@@ -240,7 +239,7 @@ describe('GeminiAgent Session Resume', () => {
       }),
     );
 
-    await vi.waitFor(() => {
+    await waitFor(() => {
       // User message
       expect(mockConnection.sessionUpdate).toHaveBeenCalledWith(
         expect.objectContaining({

packages/cli/src/acp/acpRpcDispatcher.test.ts

@@ -0,0 +1,338 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import {
+  describe,
+  it,
+  expect,
+  vi,
+  beforeEach,
+  type Mock,
+  type Mocked,
+} from 'vitest';
+import { GeminiAgent } from './acpRpcDispatcher.js';
+import * as acp from '@agentclientprotocol/sdk';
+import {
+  AuthType,
+  type Config,
+  type MessageBus,
+  type Storage,
+} from '@google/gemini-cli-core';
+import type { LoadedSettings } from '../config/settings.js';
+import { loadCliConfig, type CliArgs } from '../config/config.js';
+import { loadSettings, SettingScope } from '../config/settings.js';
+
+vi.mock('../config/config.js', () => ({
+  loadCliConfig: vi.fn(),
+}));
+
+vi.mock('../config/settings.js', async (importOriginal) => {
+  const actual = await importOriginal<typeof import('../config/settings.js')>();
+  return {
+    ...actual,
+    loadSettings: vi.fn(),
+  };
+});
+
+describe('GeminiAgent - RPC Dispatcher', () => {
+  let mockConfig: Mocked<Config>;
+  let mockSettings: Mocked<LoadedSettings>;
+  let mockArgv: CliArgs;
+  let mockConnection: Mocked<acp.AgentSideConnection>;
+  let agent: GeminiAgent;
+
+  beforeEach(() => {
+    mockConfig = {
+      refreshAuth: vi.fn(),
+      initialize: vi.fn(),
+      waitForMcpInit: vi.fn(),
+      getFileSystemService: vi.fn(),
+      setFileSystemService: vi.fn(),
+      getContentGeneratorConfig: vi.fn(),
+      getActiveModel: vi.fn().mockReturnValue('gemini-pro'),
+      getModel: vi.fn().mockReturnValue('gemini-pro'),
+      getGeminiClient: vi.fn().mockReturnValue({
+        startChat: vi.fn().mockResolvedValue({}),
+      }),
+      getMessageBus: vi.fn().mockReturnValue({
+        publish: vi.fn(),
+        subscribe: vi.fn(),
+        unsubscribe: vi.fn(),
+      }),
+      getApprovalMode: vi.fn().mockReturnValue('default'),
+      isPlanEnabled: vi.fn().mockReturnValue(true),
+      getGemini31LaunchedSync: vi.fn().mockReturnValue(false),
+      getHasAccessToPreviewModel: vi.fn().mockReturnValue(false),
+      getCheckpointingEnabled: vi.fn().mockReturnValue(false),
+      getDisableAlwaysAllow: vi.fn().mockReturnValue(false),
+      validatePathAccess: vi.fn().mockReturnValue(null),
+      getWorkspaceContext: vi.fn().mockReturnValue({
+        addReadOnlyPath: vi.fn(),
+      }),
+      getPolicyEngine: vi.fn().mockReturnValue({
+        addRule: vi.fn(),
+      }),
+      messageBus: {
+        publish: vi.fn(),
+        subscribe: vi.fn(),
+        unsubscribe: vi.fn(),
+      } as unknown as MessageBus,
+      storage: {
+        getWorkspaceAutoSavedPolicyPath: vi.fn(),
+        getAutoSavedPolicyPath: vi.fn(),
+      } as unknown as Storage,
+
+      get config() {
+        return this;
+      },
+    } as unknown as Mocked<Config>;
+    mockSettings = {
+      merged: {
+        security: { auth: { selectedType: 'login_with_google' } },
+        mcpServers: {},
+      },
+      setValue: vi.fn(),
+    } as unknown as Mocked<LoadedSettings>;
+    mockArgv = {} as unknown as CliArgs;
+    mockConnection = {
+      sessionUpdate: vi.fn(),
+      requestPermission: vi.fn(),
+    } as unknown as Mocked<acp.AgentSideConnection>;
+
+    (loadCliConfig as unknown as Mock).mockResolvedValue(mockConfig);
+    (loadSettings as unknown as Mock).mockImplementation(() => ({
+      merged: {
+        security: {
+          auth: { selectedType: AuthType.LOGIN_WITH_GOOGLE },
+          enablePermanentToolApproval: true,
+        },
+        mcpServers: {},
+      },
+      setValue: vi.fn(),
+    }));
+
+    agent = new GeminiAgent(mockConfig, mockSettings, mockArgv, mockConnection);
+  });
+
+  it('should initialize correctly', async () => {
+    const response = await agent.initialize({
+      clientCapabilities: { fs: { readTextFile: true, writeTextFile: true } },
+      protocolVersion: 1,
+    });
+
+    expect(response.protocolVersion).toBe(acp.PROTOCOL_VERSION);
+    expect(response.authMethods).toHaveLength(4);
+    const gatewayAuth = response.authMethods?.find(
+      (m) => m.id === AuthType.GATEWAY,
+    );
+    expect(gatewayAuth?._meta).toEqual({
+      gateway: {
+        protocol: 'google',
+        restartRequired: 'false',
+      },
+    });
+    const geminiAuth = response.authMethods?.find(
+      (m) => m.id === AuthType.USE_GEMINI,
+    );
+    expect(geminiAuth?._meta).toEqual({
+      'api-key': {
+        provider: 'google',
+      },
+    });
+    expect(response.agentCapabilities?.loadSession).toBe(true);
+  });
+
+  it('should authenticate correctly', async () => {
+    await agent.authenticate({
+      methodId: AuthType.LOGIN_WITH_GOOGLE,
+    });
+
+    expect(mockConfig.refreshAuth).toHaveBeenCalledWith(
+      AuthType.LOGIN_WITH_GOOGLE,
+      undefined,
+      undefined,
+      undefined,
+    );
+    expect(mockSettings.setValue).toHaveBeenCalledWith(
+      SettingScope.User,
+      'security.auth.selectedType',
+      AuthType.LOGIN_WITH_GOOGLE,
+    );
+  });
+
+  it('should authenticate correctly with api-key in _meta', async () => {
+    await agent.authenticate({
+      methodId: AuthType.USE_GEMINI,
+      _meta: {
+        'api-key': 'test-api-key',
+      },
+    } as unknown as acp.AuthenticateRequest);
+
+    expect(mockConfig.refreshAuth).toHaveBeenCalledWith(
+      AuthType.USE_GEMINI,
+      'test-api-key',
+      undefined,
+      undefined,
+    );
+    expect(mockSettings.setValue).toHaveBeenCalledWith(
+      SettingScope.User,
+      'security.auth.selectedType',
+      AuthType.USE_GEMINI,
+    );
+  });
+
+  it('should authenticate correctly with gateway method', async () => {
+    await agent.authenticate({
+      methodId: AuthType.GATEWAY,
+      _meta: {
+        gateway: {
+          baseUrl: 'https://example.com',
+          headers: { Authorization: 'Bearer token' },
+        },
+      },
+    } as unknown as acp.AuthenticateRequest);
+
+    expect(mockConfig.refreshAuth).toHaveBeenCalledWith(
+      AuthType.GATEWAY,
+      undefined,
+      'https://example.com',
+      { Authorization: 'Bearer token' },
+    );
+    expect(mockSettings.setValue).toHaveBeenCalledWith(
+      SettingScope.User,
+      'security.auth.selectedType',
+      AuthType.GATEWAY,
+    );
+  });
+
+  it('should throw acp.RequestError when gateway payload is malformed', async () => {
+    await expect(
+      agent.authenticate({
+        methodId: AuthType.GATEWAY,
+        _meta: {
+          gateway: {
+            baseUrl: 123,
+            headers: { Authorization: 'Bearer token' },
+          },
+        },
+      } as unknown as acp.AuthenticateRequest),
+    ).rejects.toThrow(/Malformed gateway payload/);
+  });
+
+  it('should cancel a session', async () => {
+    const mockSession = {
+      cancelPendingPrompt: vi.fn(),
+    };
+    (
+      agent as unknown as { sessionManager: { getSession: Mock } }
+    ).sessionManager = {
+      getSession: vi.fn().mockReturnValue(mockSession),
+    };
+
+    await agent.cancel({ sessionId: 'test-session-id' });
+
+    expect(mockSession.cancelPendingPrompt).toHaveBeenCalled();
+  });
+
+  it('should throw error when cancelling non-existent session', async () => {
+    (
+      agent as unknown as { sessionManager: { getSession: Mock } }
+    ).sessionManager = {
+      getSession: vi.fn().mockReturnValue(undefined),
+    };
+
+    await expect(agent.cancel({ sessionId: 'unknown' })).rejects.toThrow(
+      'Session not found',
+    );
+  });
+
+  it('should delegate prompt to session', async () => {
+    const mockSession = {
+      prompt: vi.fn().mockResolvedValue({ stopReason: 'end_turn' }),
+    };
+    (
+      agent as unknown as { sessionManager: { getSession: Mock } }
+    ).sessionManager = {
+      getSession: vi.fn().mockReturnValue(mockSession),
+    };
+
+    const result = await agent.prompt({
+      sessionId: 'test-session-id',
+      prompt: [],
+    });
+
+    expect(mockSession.prompt).toHaveBeenCalled();
+    expect(result).toMatchObject({ stopReason: 'end_turn' });
+  });
+
+  it('should delegate setMode to session', async () => {
+    const mockSession = {
+      setMode: vi.fn().mockReturnValue({}),
+    };
+    (
+      agent as unknown as { sessionManager: { getSession: Mock } }
+    ).sessionManager = {
+      getSession: vi.fn().mockReturnValue(mockSession),
+    };
+
+    const result = await agent.setSessionMode({
+      sessionId: 'test-session-id',
+      modeId: 'plan',
+    });
+
+    expect(mockSession.setMode).toHaveBeenCalledWith('plan');
+    expect(result).toEqual({});
+  });
+
+  it('should throw error when setting mode on non-existent session', async () => {
+    (
+      agent as unknown as { sessionManager: { getSession: Mock } }
+    ).sessionManager = {
+      getSession: vi.fn().mockReturnValue(undefined),
+    };
+
+    await expect(
+      agent.setSessionMode({
+        sessionId: 'unknown',
+        modeId: 'plan',
+      }),
+    ).rejects.toThrow('Session not found: unknown');
+  });
+
+  it('should delegate setModel to session (unstable)', async () => {
+    const mockSession = {
+      setModel: vi.fn().mockReturnValue({}),
+    };
+    (
+      agent as unknown as { sessionManager: { getSession: Mock } }
+    ).sessionManager = {
+      getSession: vi.fn().mockReturnValue(mockSession),
+    };
+
+    const result = await agent.unstable_setSessionModel({
+      sessionId: 'test-session-id',
+      modelId: 'gemini-2.0-pro-exp',
+    });
+
+    expect(mockSession.setModel).toHaveBeenCalledWith('gemini-2.0-pro-exp');
+    expect(result).toEqual({});
+  });
+
+  it('should throw error when setting model on non-existent session (unstable)', async () => {
+    (
+      agent as unknown as { sessionManager: { getSession: Mock } }
+    ).sessionManager = {
+      getSession: vi.fn().mockReturnValue(undefined),
+    };
+
+    await expect(
+      agent.unstable_setSessionModel({
+        sessionId: 'unknown',
+        modelId: 'gemini-2.0-pro-exp',
+      }),
+    ).rejects.toThrow('Session not found: unknown');
+  });
+});

packages/cli/src/acp/acpRpcDispatcher.ts

@@ -0,0 +1,232 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import {
+  type AgentLoopContext,
+  AuthType,
+  clearCachedCredentialFile,
+  getVersion,
+} from '@google/gemini-cli-core';
+import * as acp from '@agentclientprotocol/sdk';
+import { z } from 'zod';
+import { SettingScope, type LoadedSettings } from '../config/settings.js';
+import type { CliArgs } from '../config/config.js';
+import { getAcpErrorMessage } from './acpErrors.js';
+import { AcpSessionManager, type AuthDetails } from './acpSessionManager.js';
+import { hasMeta } from './acpUtils.js';
+
+export class GeminiAgent {
+  private apiKey: string | undefined;
+  private baseUrl: string | undefined;
+  private customHeaders: Record<string, string> | undefined;
+  private sessionManager: AcpSessionManager;
+
+  constructor(
+    private context: AgentLoopContext,
+    private settings: LoadedSettings,
+    argv: CliArgs,
+    connection: acp.AgentSideConnection,
+  ) {
+    this.sessionManager = new AcpSessionManager(settings, argv, connection);
+  }
+
+  async initialize(
+    args: acp.InitializeRequest,
+  ): Promise<acp.InitializeResponse> {
+    if (args.clientCapabilities) {
+      this.sessionManager.setClientCapabilities(args.clientCapabilities);
+    }
+
+    const authMethods = [
+      {
+        id: AuthType.LOGIN_WITH_GOOGLE,
+        name: 'Log in with Google',
+        description: 'Log in with your Google account',
+      },
+      {
+        id: AuthType.USE_GEMINI,
+        name: 'Gemini API key',
+        description: 'Use an API key with Gemini Developer API',
+        _meta: {
+          'api-key': {
+            provider: 'google',
+          },
+        },
+      },
+      {
+        id: AuthType.USE_VERTEX_AI,
+        name: 'Vertex AI',
+        description: 'Use an API key with Vertex AI GenAI API',
+      },
+      {
+        id: AuthType.GATEWAY,
+        name: 'AI API Gateway',
+        description: 'Use a custom AI API Gateway',
+        _meta: {
+          gateway: {
+            protocol: 'google',
+            restartRequired: 'false',
+          },
+        },
+      },
+    ];
+
+    await this.context.config.initialize();
+    const version = await getVersion();
+    return {
+      protocolVersion: acp.PROTOCOL_VERSION,
+      authMethods,
+      agentInfo: {
+        name: 'gemini-cli',
+        title: 'Gemini CLI',
+        version,
+      },
+      agentCapabilities: {
+        loadSession: true,
+        promptCapabilities: {
+          image: true,
+          audio: true,
+          embeddedContext: true,
+        },
+        mcpCapabilities: {
+          http: true,
+          sse: true,
+        },
+      },
+    };
+  }
+
+  async authenticate(req: acp.AuthenticateRequest): Promise<void> {
+    const { methodId } = req;
+    const method = z.nativeEnum(AuthType).parse(methodId);
+    const selectedAuthType = this.settings.merged.security.auth.selectedType;
+
+    // Only clear credentials when switching to a different auth method
+    if (selectedAuthType && selectedAuthType !== method) {
+      await clearCachedCredentialFile();
+    }
+    // Check for api-key in _meta
+    const meta = hasMeta(req) ? req._meta : undefined;
+    const apiKey =
+      typeof meta?.['api-key'] === 'string' ? meta['api-key'] : undefined;
+
+    // Refresh auth with the requested method
+    // This will reuse existing credentials if they're valid,
+    // or perform new authentication if needed
+    try {
+      if (apiKey) {
+        this.apiKey = apiKey;
+      }
+
+      // Extract gateway details if present
+      const gatewaySchema = z.object({
+        baseUrl: z.string().optional(),
+        headers: z.record(z.string()).optional(),
+      });
+
+      let baseUrl: string | undefined;
+      let headers: Record<string, string> | undefined;
+
+      if (meta?.['gateway']) {
+        const result = gatewaySchema.safeParse(meta['gateway']);
+        if (result.success) {
+          baseUrl = result.data.baseUrl;
+          headers = result.data.headers;
+        } else {
+          throw new acp.RequestError(
+            -32602,
+            `Malformed gateway payload: ${result.error.message}`,
+          );
+        }
+      }
+
+      this.baseUrl = baseUrl;
+      this.customHeaders = headers;
+
+      await this.context.config.refreshAuth(
+        method,
+        apiKey ?? this.apiKey,
+        baseUrl,
+        headers,
+      );
+    } catch (e) {
+      throw new acp.RequestError(-32000, getAcpErrorMessage(e));
+    }
+    this.settings.setValue(
+      SettingScope.User,
+      'security.auth.selectedType',
+      method,
+    );
+  }
+
+  private getAuthDetails(): AuthDetails {
+    return {
+      apiKey: this.apiKey,
+      baseUrl: this.baseUrl,
+      customHeaders: this.customHeaders,
+    };
+  }
+
+  async newSession(
+    params: acp.NewSessionRequest,
+  ): Promise<acp.NewSessionResponse> {
+    return this.sessionManager.newSession(params, this.getAuthDetails());
+  }
+
+  async loadSession(
+    params: acp.LoadSessionRequest,
+  ): Promise<acp.LoadSessionResponse> {
+    return this.sessionManager.loadSession(params, this.getAuthDetails());
+  }
+
+  async cancel(params: acp.CancelNotification): Promise<void> {
+    const session = this.sessionManager.getSession(params.sessionId);
+    if (!session) {
+      throw new acp.RequestError(
+        -32602,
+        `Session not found: ${params.sessionId}`,
+      );
+    }
+    await session.cancelPendingPrompt();
+  }
+
+  async prompt(params: acp.PromptRequest): Promise<acp.PromptResponse> {
+    const session = this.sessionManager.getSession(params.sessionId);
+    if (!session) {
+      throw new acp.RequestError(
+        -32602,
+        `Session not found: ${params.sessionId}`,
+      );
+    }
+    return session.prompt(params);
+  }
+
+  async setSessionMode(
+    params: acp.SetSessionModeRequest,
+  ): Promise<acp.SetSessionModeResponse> {
+    const session = this.sessionManager.getSession(params.sessionId);
+    if (!session) {
+      throw new acp.RequestError(
+        -32602,
+        `Session not found: ${params.sessionId}`,
+      );
+    }
+    return session.setMode(params.modeId);
+  }
+
+  async unstable_setSessionModel(
+    params: acp.SetSessionModelRequest,
+  ): Promise<acp.SetSessionModelResponse> {
+    const session = this.sessionManager.getSession(params.sessionId);
+    if (!session) {
+      throw new acp.RequestError(
+        -32602,
+        `Session not found: ${params.sessionId}`,
+      );
+    }
+    return session.setModel(params.modelId);
+  }
+}

packages/cli/src/acp/acpSession.test.ts

@@ -0,0 +1,463 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import {
+  describe,
+  it,
+  expect,
+  vi,
+  beforeEach,
+  afterEach,
+  type Mock,
+  type Mocked,
+} from 'vitest';
+import { Session } from './acpSession.js';
+import type * as acp from '@agentclientprotocol/sdk';
+import {
+  StreamEventType,
+  ReadManyFilesTool,
+  type GeminiChat,
+  type Config,
+  type MessageBus,
+  LlmRole,
+  type GitService,
+  type ModelRouterService,
+  InvalidStreamError,
+} from '@google/gemini-cli-core';
+import type { LoadedSettings } from '../config/settings.js';
+import * as fs from 'node:fs/promises';
+import * as path from 'node:path';
+import type { CommandHandler } from './acpCommandHandler.js';
+
+vi.mock('node:fs/promises');
+vi.mock('node:path', async (importOriginal) => {
+  const actual = await importOriginal<typeof import('node:path')>();
+  return {
+    ...actual,
+    resolve: vi.fn(),
+  };
+});
+
+vi.mock(
+  '@google/gemini-cli-core',
+  async (
+    importOriginal: () => Promise<typeof import('@google/gemini-cli-core')>,
+  ) => {
+    const actual = await importOriginal();
+    return {
+      ...actual,
+      updatePolicy: vi.fn(),
+      ReadManyFilesTool: vi.fn(),
+      logToolCall: vi.fn(),
+      processSingleFileContent: vi.fn(),
+    };
+  },
+);
+
+// eslint-disable-next-line @typescript-eslint/no-explicit-any
+async function* createMockStream(items: any[]) {
+  for (const item of items) {
+    yield item;
+  }
+}
+
+describe('Session', () => {
+  let mockChat: Mocked<GeminiChat>;
+  let mockConfig: Mocked<Config>;
+  let mockConnection: Mocked<acp.AgentSideConnection>;
+  let session: Session;
+  let mockToolRegistry: { getTool: Mock };
+  let mockTool: { kind: string; build: Mock };
+  let mockMessageBus: Mocked<MessageBus>;
+
+  beforeEach(() => {
+    mockChat = {
+      sendMessageStream: vi.fn(),
+      addHistory: vi.fn(),
+      recordCompletedToolCalls: vi.fn(),
+      getHistory: vi.fn().mockReturnValue([]),
+    } as unknown as Mocked<GeminiChat>;
+    mockTool = {
+      kind: 'read',
+      build: vi.fn().mockReturnValue({
+        getDescription: () => 'Test Tool',
+        toolLocations: () => [],
+        shouldConfirmExecute: vi.fn().mockResolvedValue(null),
+        execute: vi.fn().mockResolvedValue({ llmContent: 'Tool Result' }),
+      }),
+    };
+    mockToolRegistry = {
+      getTool: vi.fn().mockReturnValue(mockTool),
+    };
+    mockMessageBus = {
+      publish: vi.fn(),
+      subscribe: vi.fn(),
+      unsubscribe: vi.fn(),
+    } as unknown as Mocked<MessageBus>;
+    mockConfig = {
+      getModel: vi.fn().mockReturnValue('gemini-pro'),
+      getActiveModel: vi.fn().mockReturnValue('gemini-pro'),
+      getModelRouterService: vi.fn().mockReturnValue({
+        route: vi.fn().mockResolvedValue({ model: 'resolved-model' }),
+      }),
+      getToolRegistry: vi.fn().mockReturnValue(mockToolRegistry),
+      getFileService: vi.fn().mockReturnValue({
+        shouldIgnoreFile: vi.fn().mockReturnValue(false),
+      }),
+      getFileFilteringOptions: vi.fn().mockReturnValue({}),
+      getFileSystemService: vi.fn().mockReturnValue({}),
+      getTargetDir: vi.fn().mockReturnValue('/tmp'),
+      getEnableRecursiveFileSearch: vi.fn().mockReturnValue(false),
+      getDebugMode: vi.fn().mockReturnValue(false),
+      getMessageBus: vi.fn().mockReturnValue(mockMessageBus),
+      setApprovalMode: vi.fn(),
+      setModel: vi.fn(),
+      isPlanEnabled: vi.fn().mockReturnValue(true),
+      getCheckpointingEnabled: vi.fn().mockReturnValue(false),
+      getGitService: vi.fn().mockResolvedValue({} as GitService),
+      validatePathAccess: vi.fn().mockReturnValue(null),
+      getWorkspaceContext: vi.fn().mockReturnValue({
+        addReadOnlyPath: vi.fn(),
+      }),
+      waitForMcpInit: vi.fn(),
+      getDisableAlwaysAllow: vi.fn().mockReturnValue(false),
+      get config() {
+        return this;
+      },
+      get toolRegistry() {
+        return mockToolRegistry;
+      },
+    } as unknown as Mocked<Config>;
+    mockConnection = {
+      sessionUpdate: vi.fn(),
+      requestPermission: vi.fn(),
+    } as unknown as Mocked<acp.AgentSideConnection>;
+
+    session = new Session('session-1', mockChat, mockConfig, mockConnection, {
+      merged: {
+        security: { enablePermanentToolApproval: true },
+        mcpServers: {},
+      },
+      errors: [],
+    } as unknown as LoadedSettings);
+
+    (ReadManyFilesTool as unknown as Mock).mockImplementation(() => ({
+      name: 'read_many_files',
+      kind: 'read',
+      build: vi.fn().mockReturnValue({
+        getDescription: () => 'Read files',
+        toolLocations: () => [],
+        execute: vi.fn().mockResolvedValue({
+          llmContent: ['--- file.txt ---\n\nFile content\n\n'],
+        }),
+      }),
+    }));
+  });
+
+  afterEach(() => {
+    vi.restoreAllMocks();
+  });
+
+  it('should send available commands', async () => {
+    await session.sendAvailableCommands();
+
+    expect(mockConnection.sessionUpdate).toHaveBeenCalledWith(
+      expect.objectContaining({
+        update: expect.objectContaining({
+          sessionUpdate: 'available_commands_update',
+        }),
+      }),
+    );
+  });
+
+  it('should await MCP initialization before processing a prompt', async () => {
+    const stream = createMockStream([
+      {
+        type: StreamEventType.CHUNK,
+        value: { candidates: [{ content: { parts: [{ text: 'Hi' }] } }] },
+      },
+    ]);
+    mockChat.sendMessageStream.mockResolvedValue(stream);
+
+    await session.prompt({
+      sessionId: 'session-1',
+      prompt: [{ type: 'text', text: 'test' }],
+    });
+
+    expect(mockConfig.waitForMcpInit).toHaveBeenCalledOnce();
+  });
+
+  it('should handle prompt with text response', async () => {
+    const stream = createMockStream([
+      {
+        type: StreamEventType.CHUNK,
+        value: {
+          candidates: [{ content: { parts: [{ text: 'Hello' }] } }],
+        },
+      },
+    ]);
+    mockChat.sendMessageStream.mockResolvedValue(stream);
+
+    const result = await session.prompt({
+      sessionId: 'session-1',
+      prompt: [{ type: 'text', text: 'Hi' }],
+    });
+
+    expect(mockChat.sendMessageStream).toHaveBeenCalled();
+    expect(mockConnection.sessionUpdate).toHaveBeenCalledWith({
+      sessionId: 'session-1',
+      update: {
+        sessionUpdate: 'agent_message_chunk',
+        content: { type: 'text', text: 'Hello' },
+      },
+    });
+    expect(result).toMatchObject({ stopReason: 'end_turn' });
+  });
+
+  it('should use model router to determine model', async () => {
+    const mockRouter = {
+      route: vi.fn().mockResolvedValue({ model: 'routed-model' }),
+    } as unknown as ModelRouterService;
+    mockConfig.getModelRouterService.mockReturnValue(mockRouter);
+
+    const stream = createMockStream([
+      {
+        type: StreamEventType.CHUNK,
+        value: {
+          candidates: [{ content: { parts: [{ text: 'Hello' }] } }],
+        },
+      },
+    ]);
+    mockChat.sendMessageStream.mockResolvedValue(stream);
+
+    await session.prompt({
+      sessionId: 'session-1',
+      prompt: [{ type: 'text', text: 'Hi' }],
+    });
+
+    expect(mockRouter.route).toHaveBeenCalled();
+    expect(mockChat.sendMessageStream).toHaveBeenCalledWith(
+      expect.objectContaining({ model: 'routed-model' }),
+      expect.any(Array),
+      expect.any(String),
+      expect.any(Object),
+      expect.any(String),
+    );
+  });
+
+  it('should handle prompt with empty response (InvalidStreamError)', async () => {
+    mockChat.sendMessageStream.mockRejectedValue(
+      new InvalidStreamError('Empty response', 'NO_RESPONSE_TEXT'),
+    );
+
+    const result = await session.prompt({
+      sessionId: 'session-1',
+      prompt: [{ type: 'text', text: 'Hi' }],
+    });
+
+    expect(result).toMatchObject({ stopReason: 'end_turn' });
+  });
+
+  it('should handle prompt with no finish reason (InvalidStreamError)', async () => {
+    mockChat.sendMessageStream.mockRejectedValue(
+      new InvalidStreamError('No finish reason', 'NO_FINISH_REASON'),
+    );
+
+    const result = await session.prompt({
+      sessionId: 'session-1',
+      prompt: [{ type: 'text', text: 'Hi' }],
+    });
+
+    expect(result).toMatchObject({ stopReason: 'end_turn' });
+  });
+
+  it('should handle /memory command', async () => {
+    const handleCommandSpy = vi
+      .spyOn(
+        (session as unknown as { commandHandler: CommandHandler })
+          .commandHandler,
+        'handleCommand',
+      )
+      .mockResolvedValue(true);
+
+    const result = await session.prompt({
+      sessionId: 'session-1',
+      prompt: [{ type: 'text', text: '/memory view' }],
+    });
+
+    expect(result).toMatchObject({ stopReason: 'end_turn' });
+    expect(handleCommandSpy).toHaveBeenCalledWith(
+      '/memory view',
+      expect.any(Object),
+    );
+  });
+
+  it('should handle tool calls', async () => {
+    const stream1 = createMockStream([
+      {
+        type: StreamEventType.CHUNK,
+        value: {
+          functionCalls: [{ name: 'test_tool', args: { foo: 'bar' } }],
+        },
+      },
+    ]);
+    const stream2 = createMockStream([
+      {
+        type: StreamEventType.CHUNK,
+        value: {
+          candidates: [{ content: { parts: [{ text: 'Result' }] } }],
+        },
+      },
+    ]);
+
+    mockChat.sendMessageStream
+      .mockResolvedValueOnce(stream1)
+      .mockResolvedValueOnce(stream2);
+
+    const result = await session.prompt({
+      sessionId: 'session-1',
+      prompt: [{ type: 'text', text: 'Call tool' }],
+    });
+
+    expect(mockToolRegistry.getTool).toHaveBeenCalledWith('test_tool');
+    expect(result).toMatchObject({ stopReason: 'end_turn' });
+  });
+
+  it('should handle tool call permission request', async () => {
+    const confirmationDetails = {
+      type: 'info',
+      onConfirm: vi.fn(),
+    };
+    mockTool.build.mockReturnValue({
+      getDescription: () => 'Test Tool',
+      toolLocations: () => [],
+      shouldConfirmExecute: vi.fn().mockResolvedValue(confirmationDetails),
+      execute: vi.fn().mockResolvedValue({ llmContent: 'Tool Result' }),
+    });
+
+    mockConnection.requestPermission.mockResolvedValue({
+      outcome: {
+        outcome: 'selected',
+        optionId: 'proceed_once',
+      },
+    });
+
+    const stream1 = createMockStream([
+      {
+        type: StreamEventType.CHUNK,
+        value: {
+          functionCalls: [{ name: 'test_tool', args: {} }],
+        },
+      },
+    ]);
+    const stream2 = createMockStream([
+      {
+        type: StreamEventType.CHUNK,
+        value: { candidates: [] },
+      },
+    ]);
+
+    mockChat.sendMessageStream
+      .mockResolvedValueOnce(stream1)
+      .mockResolvedValueOnce(stream2);
+
+    await session.prompt({
+      sessionId: 'session-1',
+      prompt: [{ type: 'text', text: 'Call tool' }],
+    });
+
+    expect(mockConnection.requestPermission).toHaveBeenCalled();
+    expect(confirmationDetails.onConfirm).toHaveBeenCalled();
+  });
+
+  it('should handle @path resolution', async () => {
+    (path.resolve as unknown as Mock).mockReturnValue('/tmp/file.txt');
+    (fs.stat as unknown as Mock).mockResolvedValue({
+      isDirectory: () => false,
+    });
+
+    const stream = createMockStream([
+      {
+        type: StreamEventType.CHUNK,
+        value: { candidates: [] },
+      },
+    ]);
+    mockChat.sendMessageStream.mockResolvedValue(stream);
+
+    await session.prompt({
+      sessionId: 'session-1',
+      prompt: [
+        { type: 'text', text: 'Read' },
+        {
+          type: 'resource_link',
+          uri: 'file://file.txt',
+          mimeType: 'text/plain',
+          name: 'file.txt',
+        },
+      ],
+    });
+
+    expect(path.resolve).toHaveBeenCalled();
+    expect(fs.stat).toHaveBeenCalled();
+    expect(mockChat.sendMessageStream).toHaveBeenCalledWith(
+      expect.anything(),
+      expect.arrayContaining([
+        expect.objectContaining({
+          text: expect.stringContaining('Content from @file.txt'),
+        }),
+      ]),
+      expect.anything(),
+      expect.any(AbortSignal),
+      LlmRole.MAIN,
+    );
+  });
+
+  it('should handle rate limit error', async () => {
+    const error = new Error('Rate limit');
+    (error as unknown as { status: number }).status = 429;
+    mockChat.sendMessageStream.mockRejectedValue(error);
+
+    await expect(
+      session.prompt({
+        sessionId: 'session-1',
+        prompt: [{ type: 'text', text: 'Hi' }],
+      }),
+    ).rejects.toMatchObject({
+      code: 429,
+      message: 'Rate limit exceeded. Try again later.',
+    });
+  });
+
+  it('should handle missing tool', async () => {
+    mockToolRegistry.getTool.mockReturnValue(undefined);
+
+    const stream1 = createMockStream([
+      {
+        type: StreamEventType.CHUNK,
+        value: {
+          functionCalls: [{ name: 'unknown_tool', args: {} }],
+        },
+      },
+    ]);
+    const stream2 = createMockStream([
+      {
+        type: StreamEventType.CHUNK,
+        value: { candidates: [] },
+      },
+    ]);
+
+    mockChat.sendMessageStream
+      .mockResolvedValueOnce(stream1)
+      .mockResolvedValueOnce(stream2);
+
+    await session.prompt({
+      sessionId: 'session-1',
+      prompt: [{ type: 'text', text: 'Call tool' }],
+    });
+
+    expect(mockChat.sendMessageStream).toHaveBeenCalledTimes(2);
+  });
+});

packages/cli/src/acp/acpSession.ts

@@ -1,27 +1,20 @@
 /**
  * @license
- * Copyright 2025 Google LLC
+ * Copyright 2026 Google LLC
  * SPDX-License-Identifier: Apache-2.0
  */
 
 import {
-  type Config,
+  type ApprovalMode,
   type GeminiChat,
   type ToolResult,
-  type ToolCallConfirmationDetails,
-  type FilterFilesOptions,
   type ConversationRecord,
   CoreToolCallStatus,
-  AuthType,
   logToolCall,
   convertToFunctionResponse,
   ToolConfirmationOutcome,
-  clearCachedCredentialFile,
-  isNodeError,
-  getErrorMessage,
   isWithinRoot,
   getErrorStatus,
-  MCPServerConfig,
   DiscoveredMCPTool,
   StreamEventType,
   ToolCallEvent,
@@ -29,547 +22,42 @@ import {
   ReadManyFilesTool,
   REFERENCE_CONTENT_START,
   type RoutingContext,
-  createWorkingStdio,
-  startupProfiler,
-  Kind,
   partListUnionToString,
   LlmRole,
-  ApprovalMode,
-  getVersion,
-  convertSessionToClientHistory,
-  DEFAULT_GEMINI_MODEL,
-  DEFAULT_GEMINI_FLASH_MODEL,
-  DEFAULT_GEMINI_FLASH_LITE_MODEL,
-  PREVIEW_GEMINI_MODEL,
-  PREVIEW_GEMINI_3_1_MODEL,
-  PREVIEW_GEMINI_3_1_FLASH_LITE_MODEL,
-  PREVIEW_GEMINI_3_1_CUSTOM_TOOLS_MODEL,
-  PREVIEW_GEMINI_FLASH_MODEL,
-  DEFAULT_GEMINI_MODEL_AUTO,
-  PREVIEW_GEMINI_MODEL_AUTO,
-  getDisplayString,
   processSingleFileContent,
   InvalidStreamError,
   type AgentLoopContext,
   updatePolicy,
+  isNodeError,
+  getErrorMessage,
+  type FilterFilesOptions,
+  isTextPart,
 } from '@google/gemini-cli-core';
 import * as acp from '@agentclientprotocol/sdk';
-import { AcpFileSystemService } from './fileSystemService.js';
-import { getAcpErrorMessage } from './acpErrors.js';
-import { Readable, Writable } from 'node:stream';
-
-function hasMeta(obj: unknown): obj is { _meta?: Record<string, unknown> } {
-  return typeof obj === 'object' && obj !== null && '_meta' in obj;
-}
 import type { Content, Part, FunctionCall } from '@google/genai';
-import {
-  SettingScope,
-  loadSettings,
-  type LoadedSettings,
-} from '../config/settings.js';
-import { createPolicyUpdater } from '../config/policy.js';
+import type { LoadedSettings } from '../config/settings.js';
 import * as fs from 'node:fs/promises';
 import * as path from 'node:path';
-import { z } from 'zod';
-
 import { randomUUID } from 'node:crypto';
-import { loadCliConfig, type CliArgs } from '../config/config.js';
-import { runExitCleanup } from '../utils/cleanup.js';
-import { SessionSelector } from '../utils/sessionUtils.js';
-import { startAutoMemoryIfEnabled } from '../utils/autoMemory.js';
-
-import { CommandHandler } from './commandHandler.js';
-
-const RequestPermissionResponseSchema = z.object({
-  outcome: z.discriminatedUnion('outcome', [
-    z.object({ outcome: z.literal('cancelled') }),
-    z.object({
-      outcome: z.literal('selected'),
-      optionId: z.string(),
-    }),
-  ]),
-});
-
-export async function runAcpClient(
-  config: Config,
-  settings: LoadedSettings,
-  argv: CliArgs,
-) {
-  // ... (skip unchanged lines) ...
-
-  const { stdout: workingStdout } = createWorkingStdio();
-  const stdout = Writable.toWeb(workingStdout) as WritableStream;
-  // eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
-  const stdin = Readable.toWeb(process.stdin) as ReadableStream<Uint8Array>;
-
-  const stream = acp.ndJsonStream(stdout, stdin);
-  const connection = new acp.AgentSideConnection(
-    (connection) => new GeminiAgent(config, settings, argv, connection),
-    stream,
-  );
-
-  // SIGTERM/SIGINT handlers (in sdk.ts) don't fire when stdin closes.
-  // We must explicitly await the connection close to flush telemetry.
-  // Use finally() to ensure cleanup runs even on stream errors.
-  await connection.closed.finally(runExitCleanup);
-}
-
-export class GeminiAgent {
-  private static callIdCounter = 0;
-
-  static generateCallId(name: string): string {
-    return `${name}-${Date.now()}-${++GeminiAgent.callIdCounter}`;
-  }
-
-  private sessions: Map<string, Session> = new Map();
-  private clientCapabilities: acp.ClientCapabilities | undefined;
-  private apiKey: string | undefined;
-  private baseUrl: string | undefined;
-  private customHeaders: Record<string, string> | undefined;
-
-  constructor(
-    private context: AgentLoopContext,
-    private settings: LoadedSettings,
-    private argv: CliArgs,
-    private connection: acp.AgentSideConnection,
-  ) {}
-
-  async initialize(
-    args: acp.InitializeRequest,
-  ): Promise<acp.InitializeResponse> {
-    this.clientCapabilities = args.clientCapabilities;
-
-    const authMethods = [
-      {
-        id: AuthType.LOGIN_WITH_GOOGLE,
-        name: 'Log in with Google',
-        description: 'Log in with your Google account',
-      },
-      {
-        id: AuthType.USE_GEMINI,
-        name: 'Gemini API key',
-        description: 'Use an API key with Gemini Developer API',
-        _meta: {
-          'api-key': {
-            provider: 'google',
-          },
-        },
-      },
-      {
-        id: AuthType.USE_VERTEX_AI,
-        name: 'Vertex AI',
-        description: 'Use an API key with Vertex AI GenAI API',
-      },
-      {
-        id: AuthType.GATEWAY,
-        name: 'AI API Gateway',
-        description: 'Use a custom AI API Gateway',
-        _meta: {
-          gateway: {
-            protocol: 'google',
-            restartRequired: 'false',
-          },
-        },
-      },
-    ];
-
-    await this.context.config.initialize();
-    const version = await getVersion();
-    return {
-      protocolVersion: acp.PROTOCOL_VERSION,
-      authMethods,
-      agentInfo: {
-        name: 'gemini-cli',
-        title: 'Gemini CLI',
-        version,
-      },
-      agentCapabilities: {
-        loadSession: true,
-        promptCapabilities: {
-          image: true,
-          audio: true,
-          embeddedContext: true,
-        },
-        mcpCapabilities: {
-          http: true,
-          sse: true,
-        },
-      },
-    };
-  }
-
-  async authenticate(req: acp.AuthenticateRequest): Promise<void> {
-    const { methodId } = req;
-    const method = z.nativeEnum(AuthType).parse(methodId);
-    const selectedAuthType = this.settings.merged.security.auth.selectedType;
-
-    // Only clear credentials when switching to a different auth method
-    if (selectedAuthType && selectedAuthType !== method) {
-      await clearCachedCredentialFile();
-    }
-    // Check for api-key in _meta
-    const meta = hasMeta(req) ? req._meta : undefined;
-    const apiKey =
-      typeof meta?.['api-key'] === 'string' ? meta['api-key'] : undefined;
-
-    // Refresh auth with the requested method
-    // This will reuse existing credentials if they're valid,
-    // or perform new authentication if needed
-    try {
-      if (apiKey) {
-        this.apiKey = apiKey;
-      }
-
-      // Extract gateway details if present
-      const gatewaySchema = z.object({
-        baseUrl: z.string().optional(),
-        headers: z.record(z.string()).optional(),
-      });
-
-      let baseUrl: string | undefined;
-      let headers: Record<string, string> | undefined;
-
-      if (meta?.['gateway']) {
-        const result = gatewaySchema.safeParse(meta['gateway']);
-        if (result.success) {
-          baseUrl = result.data.baseUrl;
-          headers = result.data.headers;
-        } else {
-          throw new acp.RequestError(
-            -32602,
-            `Malformed gateway payload: ${result.error.message}`,
-          );
-        }
-      }
-
-      this.baseUrl = baseUrl;
-      this.customHeaders = headers;
-
-      await this.context.config.refreshAuth(
-        method,
-        apiKey ?? this.apiKey,
-        baseUrl,
-        headers,
-      );
-    } catch (e) {
-      throw new acp.RequestError(-32000, getAcpErrorMessage(e));
-    }
-    this.settings.setValue(
-      SettingScope.User,
-      'security.auth.selectedType',
-      method,
-    );
-  }
-
-  async newSession({
-    cwd,
-    mcpServers,
-  }: acp.NewSessionRequest): Promise<acp.NewSessionResponse> {
-    const sessionId = randomUUID();
-    const loadedSettings = loadSettings(cwd);
-    const config = await this.newSessionConfig(
-      sessionId,
-      cwd,
-      mcpServers,
-      loadedSettings,
-    );
-
-    const authType =
-      loadedSettings.merged.security.auth.selectedType || AuthType.USE_GEMINI;
-
-    let isAuthenticated = false;
-    let authErrorMessage = '';
-    try {
-      await config.refreshAuth(
-        authType,
-        this.apiKey,
-        this.baseUrl,
-        this.customHeaders,
-      );
-      isAuthenticated = true;
-
-      // Extra validation for Gemini API key
-      const contentGeneratorConfig = config.getContentGeneratorConfig();
-      if (
-        authType === AuthType.USE_GEMINI &&
-        (!contentGeneratorConfig || !contentGeneratorConfig.apiKey)
-      ) {
-        isAuthenticated = false;
-        authErrorMessage = 'Gemini API key is missing or not configured.';
-      }
-    } catch (e) {
-      isAuthenticated = false;
-      authErrorMessage = getAcpErrorMessage(e);
-      debugLogger.error(
-        `Authentication failed: ${e instanceof Error ? e.stack : e}`,
-      );
-    }
-
-    if (!isAuthenticated) {
-      throw new acp.RequestError(
-        -32000,
-        authErrorMessage || 'Authentication required.',
-      );
-    }
-
-    if (this.clientCapabilities?.fs) {
-      const acpFileSystemService = new AcpFileSystemService(
-        this.connection,
-        sessionId,
-        this.clientCapabilities.fs,
-        config.getFileSystemService(),
-        cwd,
-      );
-      config.setFileSystemService(acpFileSystemService);
-    }
-
-    await config.initialize();
-    startupProfiler.flush(config);
-    startAutoMemoryIfEnabled(config);
-
-    const geminiClient = config.getGeminiClient();
-    const chat = await geminiClient.startChat();
-
-    const session = new Session(
-      sessionId,
-      chat,
-      config,
-      this.connection,
-      this.settings,
-    );
-    this.sessions.set(sessionId, session);
-
-    setTimeout(() => {
-      // eslint-disable-next-line @typescript-eslint/no-floating-promises
-      session.sendAvailableCommands();
-    }, 0);
-
-    const { availableModels, currentModelId } = buildAvailableModels(
-      config,
-      loadedSettings,
-    );
-
-    const response = {
-      sessionId,
-      modes: {
-        availableModes: buildAvailableModes(config.isPlanEnabled()),
-        currentModeId: config.getApprovalMode(),
-      },
-      models: {
-        availableModels,
-        currentModelId,
-      },
-    };
-    return response;
-  }
-
-  async loadSession({
-    sessionId,
-    cwd,
-    mcpServers,
-  }: acp.LoadSessionRequest): Promise<acp.LoadSessionResponse> {
-    const config = await this.initializeSessionConfig(
-      sessionId,
-      cwd,
-      mcpServers,
-    );
-
-    const sessionSelector = new SessionSelector(config.storage);
-    const { sessionData, sessionPath } =
-      await sessionSelector.resolveSession(sessionId);
-
-    const clientHistory = convertSessionToClientHistory(sessionData.messages);
-
-    const geminiClient = config.getGeminiClient();
-    await geminiClient.initialize();
-    await geminiClient.resumeChat(clientHistory, {
-      conversation: sessionData,
-      filePath: sessionPath,
-    });
-
-    const session = new Session(
-      sessionId,
-      geminiClient.getChat(),
-      config,
-      this.connection,
-      this.settings,
-    );
-    this.sessions.set(sessionId, session);
-
-    // Stream history back to client
-    // eslint-disable-next-line @typescript-eslint/no-floating-promises
-    session.streamHistory(sessionData.messages);
-
-    setTimeout(() => {
-      // eslint-disable-next-line @typescript-eslint/no-floating-promises
-      session.sendAvailableCommands();
-    }, 0);
-
-    const { availableModels, currentModelId } = buildAvailableModels(
-      config,
-      this.settings,
-    );
-
-    const response = {
-      modes: {
-        availableModes: buildAvailableModes(config.isPlanEnabled()),
-        currentModeId: config.getApprovalMode(),
-      },
-      models: {
-        availableModels,
-        currentModelId,
-      },
-    };
-    return response;
-  }
-
-  private async initializeSessionConfig(
-    sessionId: string,
-    cwd: string,
-    mcpServers: acp.McpServer[],
-  ): Promise<Config> {
-    const selectedAuthType = this.settings.merged.security.auth.selectedType;
-    if (!selectedAuthType) {
-      throw acp.RequestError.authRequired();
-    }
-
-    // 1. Create config WITHOUT initializing it (no MCP servers started yet)
-    const config = await this.newSessionConfig(sessionId, cwd, mcpServers);
-
-    // 2. Authenticate BEFORE initializing configuration or starting MCP servers.
-    // This satisfies the security requirement to verify the user before executing
-    // potentially unsafe server definitions.
-    try {
-      await config.refreshAuth(
-        selectedAuthType,
-        this.apiKey,
-        this.baseUrl,
-        this.customHeaders,
-      );
-    } catch (e) {
-      debugLogger.error(`Authentication failed: ${e}`);
-      throw acp.RequestError.authRequired();
-    }
-
-    // 3. Set the ACP FileSystemService (if supported) before config initialization
-    if (this.clientCapabilities?.fs) {
-      const acpFileSystemService = new AcpFileSystemService(
-        this.connection,
-        sessionId,
-        this.clientCapabilities.fs,
-        config.getFileSystemService(),
-        cwd,
-      );
-      config.setFileSystemService(acpFileSystemService);
-    }
-
-    // 4. Now that we are authenticated, it is safe to initialize the config
-    // which starts the MCP servers and other heavy resources.
-    await config.initialize();
-    startupProfiler.flush(config);
-    startAutoMemoryIfEnabled(config);
-
-    return config;
-  }
-
-  async newSessionConfig(
-    sessionId: string,
-    cwd: string,
-    mcpServers: acp.McpServer[],
-    loadedSettings?: LoadedSettings,
-  ): Promise<Config> {
-    const currentSettings = loadedSettings || this.settings;
-    const mergedMcpServers = { ...currentSettings.merged.mcpServers };
-
-    for (const server of mcpServers) {
-      if (
-        'type' in server &&
-        (server.type === 'sse' || server.type === 'http')
-      ) {
-        // HTTP or SSE MCP server
-        const headers = Object.fromEntries(
-          server.headers.map(({ name, value }) => [name, value]),
-        );
-        mergedMcpServers[server.name] = new MCPServerConfig(
-          undefined, // command
-          undefined, // args
-          undefined, // env
-          undefined, // cwd
-          server.type === 'sse' ? server.url : undefined, // url (sse)
-          server.type === 'http' ? server.url : undefined, // httpUrl
-          headers,
-        );
-      } else if ('command' in server) {
-        // Stdio MCP server
-        const env: Record<string, string> = {};
-        for (const { name: envName, value } of server.env) {
-          env[envName] = value;
-        }
-        mergedMcpServers[server.name] = new MCPServerConfig(
-          server.command,
-          server.args,
-          env,
-          cwd,
-        );
-      }
-    }
-
-    const settings = {
-      ...currentSettings.merged,
-      mcpServers: mergedMcpServers,
-    };
-
-    const config = await loadCliConfig(settings, sessionId, this.argv, { cwd });
-
-    createPolicyUpdater(
-      config.getPolicyEngine(),
-      config.messageBus,
-      config.storage,
-    );
-
-    return config;
-  }
-
-  async cancel(params: acp.CancelNotification): Promise<void> {
-    const session = this.sessions.get(params.sessionId);
-    if (!session) {
-      throw new Error(`Session not found: ${params.sessionId}`);
-    }
-    await session.cancelPendingPrompt();
-  }
-
-  async prompt(params: acp.PromptRequest): Promise<acp.PromptResponse> {
-    const session = this.sessions.get(params.sessionId);
-    if (!session) {
-      throw new Error(`Session not found: ${params.sessionId}`);
-    }
-    return session.prompt(params);
-  }
-
-  async setSessionMode(
-    params: acp.SetSessionModeRequest,
-  ): Promise<acp.SetSessionModeResponse> {
-    const session = this.sessions.get(params.sessionId);
-    if (!session) {
-      throw new Error(`Session not found: ${params.sessionId}`);
-    }
-    return session.setMode(params.modeId);
-  }
-
-  async unstable_setSessionModel(
-    params: acp.SetSessionModelRequest,
-  ): Promise<acp.SetSessionModelResponse> {
-    const session = this.sessions.get(params.sessionId);
-    if (!session) {
-      throw new Error(`Session not found: ${params.sessionId}`);
-    }
-    return session.setModel(params.modelId);
-  }
-}
+import { CommandHandler } from './acpCommandHandler.js';
+import {
+  toToolCallContent,
+  toPermissionOptions,
+  toAcpToolKind,
+  buildAvailableModes,
+  RequestPermissionResponseSchema,
+} from './acpUtils.js';
+import { z } from 'zod';
+import { getAcpErrorMessage } from './acpErrors.js';
 
 export class Session {
   private pendingPrompt: AbortController | null = null;
   private commandHandler = new CommandHandler();
+  private callIdCounter = 0;
+
+  private generateCallId(name: string): string {
+    return `${name}-${Date.now()}-${++this.callIdCounter}`;
+  }
 
   constructor(
     private readonly id: string,
@@ -709,13 +197,10 @@ export class Session {
 
     for (const part of parts) {
       if (typeof part === 'object' && part !== null) {
-        if ('text' in part) {
+        if (isTextPart(part)) {
           // It is a text part
-          // eslint-disable-next-line @typescript-eslint/no-explicit-any, @typescript-eslint/no-unsafe-assignment, @typescript-eslint/no-unsafe-type-assertion
-          const text = (part as any).text;
-          if (typeof text === 'string') {
-            commandText += text;
-          }
+          const text = part.text;
+          commandText += text;
         } else {
           // Non-text part (image, embedded resource)
           // Stop looking for command
@@ -972,7 +457,7 @@ export class Session {
     promptId: string,
     fc: FunctionCall,
   ): Promise<Part[]> {
-    const callId = fc.id ?? GeminiAgent.generateCallId(fc.name || 'unknown');
+    const callId = fc.id ?? this.generateCallId(fc.name || 'unknown');
     const args = fc.args ?? {};
 
     const startTime = Date.now();
@@ -1661,7 +1146,7 @@ export class Session {
         include: pathSpecsToRead,
       };
 
-      const callId = GeminiAgent.generateCallId(readManyFilesTool.name);
+      const callId = this.generateCallId(readManyFilesTool.name);
 
       try {
         const invocation = readManyFilesTool.build(toolArgs);
@@ -1799,333 +1284,3 @@ export class Session {
     }
   }
 }
-
-function toToolCallContent(toolResult: ToolResult): acp.ToolCallContent | null {
-  if (toolResult.error?.message) {
-    throw new Error(toolResult.error.message);
-  }
-
-  if (toolResult.returnDisplay) {
-    if (typeof toolResult.returnDisplay === 'string') {
-      return {
-        type: 'content',
-        content: { type: 'text', text: toolResult.returnDisplay },
-      };
-    } else {
-      if ('fileName' in toolResult.returnDisplay) {
-        return {
-          type: 'diff',
-          path:
-            toolResult.returnDisplay.filePath ??
-            toolResult.returnDisplay.fileName,
-          oldText: toolResult.returnDisplay.originalContent,
-          newText: toolResult.returnDisplay.newContent,
-          _meta: {
-            kind: !toolResult.returnDisplay.originalContent
-              ? 'add'
-              : toolResult.returnDisplay.newContent === ''
-                ? 'delete'
-                : 'modify',
-          },
-        };
-      }
-      return null;
-    }
-  } else {
-    return null;
-  }
-}
-
-const basicPermissionOptions = [
-  {
-    optionId: ToolConfirmationOutcome.ProceedOnce,
-    name: 'Allow',
-    kind: 'allow_once',
-  },
-  {
-    optionId: ToolConfirmationOutcome.Cancel,
-    name: 'Reject',
-    kind: 'reject_once',
-  },
-] as const;
-
-function toPermissionOptions(
-  confirmation: ToolCallConfirmationDetails,
-  config: Config,
-  enablePermanentToolApproval: boolean = false,
-): acp.PermissionOption[] {
-  const disableAlwaysAllow = config.getDisableAlwaysAllow();
-  const options: acp.PermissionOption[] = [];
-
-  if (!disableAlwaysAllow) {
-    switch (confirmation.type) {
-      case 'edit':
-        options.push({
-          optionId: ToolConfirmationOutcome.ProceedAlways,
-          name: 'Allow for this session',
-          kind: 'allow_always',
-        });
-        if (enablePermanentToolApproval) {
-          options.push({
-            optionId: ToolConfirmationOutcome.ProceedAlwaysAndSave,
-            name: 'Allow for this file in all future sessions',
-            kind: 'allow_always',
-          });
-        }
-        break;
-      case 'exec':
-        options.push({
-          optionId: ToolConfirmationOutcome.ProceedAlways,
-          name: 'Allow for this session',
-          kind: 'allow_always',
-        });
-        if (enablePermanentToolApproval) {
-          options.push({
-            optionId: ToolConfirmationOutcome.ProceedAlwaysAndSave,
-            name: 'Allow this command for all future sessions',
-            kind: 'allow_always',
-          });
-        }
-        break;
-      case 'mcp':
-        options.push(
-          {
-            optionId: ToolConfirmationOutcome.ProceedAlwaysServer,
-            name: 'Allow all server tools for this session',
-            kind: 'allow_always',
-          },
-          {
-            optionId: ToolConfirmationOutcome.ProceedAlwaysTool,
-            name: 'Allow tool for this session',
-            kind: 'allow_always',
-          },
-        );
-        if (enablePermanentToolApproval) {
-          options.push({
-            optionId: ToolConfirmationOutcome.ProceedAlwaysAndSave,
-            name: 'Allow tool for all future sessions',
-            kind: 'allow_always',
-          });
-        }
-        break;
-      case 'info':
-        options.push({
-          optionId: ToolConfirmationOutcome.ProceedAlways,
-          name: 'Allow for this session',
-          kind: 'allow_always',
-        });
-        if (enablePermanentToolApproval) {
-          options.push({
-            optionId: ToolConfirmationOutcome.ProceedAlwaysAndSave,
-            name: 'Allow for all future sessions',
-            kind: 'allow_always',
-          });
-        }
-        break;
-      case 'ask_user':
-      case 'exit_plan_mode':
-        // askuser and exit_plan_mode don't need "always allow" options
-        break;
-      default:
-        // No "always allow" options for other types
-        break;
-    }
-  }
-
-  options.push(...basicPermissionOptions);
-
-  // Exhaustive check
-  switch (confirmation.type) {
-    case 'edit':
-    case 'exec':
-    case 'mcp':
-    case 'info':
-    case 'ask_user':
-    case 'exit_plan_mode':
-    case 'sandbox_expansion':
-      break;
-    default: {
-      const unreachable: never = confirmation;
-      throw new Error(`Unexpected: ${unreachable}`);
-    }
-  }
-
-  return options;
-}
-
-/**
- * Maps our internal tool kind to the ACP ToolKind.
- * Fallback to 'other' for kinds that are not supported by the ACP protocol.
- */
-function toAcpToolKind(kind: Kind): acp.ToolKind {
-  switch (kind) {
-    case Kind.Read:
-    case Kind.Edit:
-    case Kind.Execute:
-    case Kind.Search:
-    case Kind.Delete:
-    case Kind.Move:
-    case Kind.Think:
-    case Kind.Fetch:
-    case Kind.SwitchMode:
-    case Kind.Other:
-      return kind as acp.ToolKind;
-    case Kind.Agent:
-      return 'think';
-    case Kind.Plan:
-    case Kind.Communicate:
-    default:
-      return 'other';
-  }
-}
-
-function buildAvailableModes(isPlanEnabled: boolean): acp.SessionMode[] {
-  const modes: acp.SessionMode[] = [
-    {
-      id: ApprovalMode.DEFAULT,
-      name: 'Default',
-      description: 'Prompts for approval',
-    },
-    {
-      id: ApprovalMode.AUTO_EDIT,
-      name: 'Auto Edit',
-      description: 'Auto-approves edit tools',
-    },
-    {
-      id: ApprovalMode.YOLO,
-      name: 'YOLO',
-      description: 'Auto-approves all tools',
-    },
-  ];
-
-  if (isPlanEnabled) {
-    modes.push({
-      id: ApprovalMode.PLAN,
-      name: 'Plan',
-      description: 'Read-only mode',
-    });
-  }
-
-  return modes;
-}
-
-function buildAvailableModels(
-  config: Config,
-  settings: LoadedSettings,
-): {
-  availableModels: Array<{
-    modelId: string;
-    name: string;
-    description?: string;
-  }>;
-  currentModelId: string;
-} {
-  const preferredModel = config.getModel() || DEFAULT_GEMINI_MODEL_AUTO;
-  const shouldShowPreviewModels = config.getHasAccessToPreviewModel();
-  const useGemini31 = config.getGemini31LaunchedSync?.() ?? false;
-  const useGemini31FlashLite =
-    config.getGemini31FlashLiteLaunchedSync?.() ?? false;
-  const selectedAuthType = settings.merged.security.auth.selectedType;
-  const useCustomToolModel =
-    useGemini31 && selectedAuthType === AuthType.USE_GEMINI;
-
-  // --- DYNAMIC PATH ---
-  if (
-    config.getExperimentalDynamicModelConfiguration?.() === true &&
-    config.getModelConfigService
-  ) {
-    const options = config.getModelConfigService().getAvailableModelOptions({
-      useGemini3_1: useGemini31,
-      useGemini3_1FlashLite: useGemini31FlashLite,
-      useCustomTools: useCustomToolModel,
-      hasAccessToPreview: shouldShowPreviewModels,
-    });
-
-    return {
-      availableModels: options,
-      currentModelId: preferredModel,
-    };
-  }
-
-  // --- LEGACY PATH ---
-  const mainOptions = [
-    {
-      value: DEFAULT_GEMINI_MODEL_AUTO,
-      title: getDisplayString(DEFAULT_GEMINI_MODEL_AUTO),
-      description:
-        'Let Gemini CLI decide the best model for the task: gemini-2.5-pro, gemini-2.5-flash',
-    },
-  ];
-
-  if (shouldShowPreviewModels) {
-    mainOptions.unshift({
-      value: PREVIEW_GEMINI_MODEL_AUTO,
-      title: getDisplayString(PREVIEW_GEMINI_MODEL_AUTO),
-      description: useGemini31
-        ? 'Let Gemini CLI decide the best model for the task: gemini-3.1-pro, gemini-3-flash'
-        : 'Let Gemini CLI decide the best model for the task: gemini-3-pro, gemini-3-flash',
-    });
-  }
-
-  const manualOptions = [
-    {
-      value: DEFAULT_GEMINI_MODEL,
-      title: getDisplayString(DEFAULT_GEMINI_MODEL),
-    },
-    {
-      value: DEFAULT_GEMINI_FLASH_MODEL,
-      title: getDisplayString(DEFAULT_GEMINI_FLASH_MODEL),
-    },
-    {
-      value: DEFAULT_GEMINI_FLASH_LITE_MODEL,
-      title: getDisplayString(DEFAULT_GEMINI_FLASH_LITE_MODEL),
-    },
-  ];
-
-  if (shouldShowPreviewModels) {
-    const previewProModel = useGemini31
-      ? PREVIEW_GEMINI_3_1_MODEL
-      : PREVIEW_GEMINI_MODEL;
-
-    const previewProValue = useCustomToolModel
-      ? PREVIEW_GEMINI_3_1_CUSTOM_TOOLS_MODEL
-      : previewProModel;
-
-    const previewOptions = [
-      {
-        value: previewProValue,
-        title: getDisplayString(previewProModel),
-      },
-      {
-        value: PREVIEW_GEMINI_FLASH_MODEL,
-        title: getDisplayString(PREVIEW_GEMINI_FLASH_MODEL),
-      },
-    ];
-
-    if (useGemini31FlashLite) {
-      previewOptions.push({
-        value: PREVIEW_GEMINI_3_1_FLASH_LITE_MODEL,
-        title: getDisplayString(PREVIEW_GEMINI_3_1_FLASH_LITE_MODEL),
-      });
-    }
-
-    manualOptions.unshift(...previewOptions);
-  }
-
-  const scaleOptions = (
-    options: Array<{ value: string; title: string; description?: string }>,
-  ) =>
-    options.map((o) => ({
-      modelId: o.value,
-      name: o.title,
-      description: o.description,
-    }));
-
-  return {
-    availableModels: [
-      ...scaleOptions(mainOptions),
-      ...scaleOptions(manualOptions),
-    ],
-    currentModelId: preferredModel,
-  };
-}

packages/cli/src/acp/acpSessionManager.test.ts

@@ -0,0 +1,386 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import {
+  describe,
+  it,
+  expect,
+  vi,
+  beforeEach,
+  afterEach,
+  type Mock,
+  type Mocked,
+} from 'vitest';
+import { AcpSessionManager } from './acpSessionManager.js';
+import type * as acp from '@agentclientprotocol/sdk';
+import {
+  AuthType,
+  type Config,
+  type MessageBus,
+  type Storage,
+} from '@google/gemini-cli-core';
+import type { LoadedSettings } from '../config/settings.js';
+import { loadCliConfig, type CliArgs } from '../config/config.js';
+import { loadSettings } from '../config/settings.js';
+
+vi.mock('../config/config.js', () => ({
+  loadCliConfig: vi.fn(),
+}));
+
+vi.mock('../config/settings.js', async (importOriginal) => {
+  const actual = await importOriginal<typeof import('../config/settings.js')>();
+  return {
+    ...actual,
+    loadSettings: vi.fn(),
+  };
+});
+
+const startAutoMemoryIfEnabledMock = vi.fn();
+vi.mock('../utils/autoMemory.js', () => ({
+  startAutoMemoryIfEnabled: (config: Config) =>
+    startAutoMemoryIfEnabledMock(config),
+}));
+
+describe('AcpSessionManager', () => {
+  let mockConfig: Mocked<Config>;
+  let mockSettings: Mocked<LoadedSettings>;
+  let mockArgv: CliArgs;
+  let mockConnection: Mocked<acp.AgentSideConnection>;
+  let manager: AcpSessionManager;
+
+  beforeEach(() => {
+    mockConfig = {
+      refreshAuth: vi.fn(),
+      initialize: vi.fn(),
+      waitForMcpInit: vi.fn(),
+      getFileSystemService: vi.fn(),
+      setFileSystemService: vi.fn(),
+      getContentGeneratorConfig: vi.fn(),
+      getActiveModel: vi.fn().mockReturnValue('gemini-pro'),
+      getModel: vi.fn().mockReturnValue('gemini-pro'),
+      getGeminiClient: vi.fn().mockReturnValue({
+        startChat: vi.fn().mockResolvedValue({}),
+      }),
+      getMessageBus: vi.fn().mockReturnValue({
+        publish: vi.fn(),
+        subscribe: vi.fn(),
+        unsubscribe: vi.fn(),
+      }),
+      getApprovalMode: vi.fn().mockReturnValue('default'),
+      isPlanEnabled: vi.fn().mockReturnValue(true),
+      getGemini31LaunchedSync: vi.fn().mockReturnValue(false),
+      getHasAccessToPreviewModel: vi.fn().mockReturnValue(false),
+      getCheckpointingEnabled: vi.fn().mockReturnValue(false),
+      getDisableAlwaysAllow: vi.fn().mockReturnValue(false),
+      validatePathAccess: vi.fn().mockReturnValue(null),
+      getWorkspaceContext: vi.fn().mockReturnValue({
+        addReadOnlyPath: vi.fn(),
+      }),
+      getPolicyEngine: vi.fn().mockReturnValue({
+        addRule: vi.fn(),
+      }),
+      messageBus: {
+        publish: vi.fn(),
+        subscribe: vi.fn(),
+        unsubscribe: vi.fn(),
+      } as unknown as MessageBus,
+      storage: {
+        getWorkspaceAutoSavedPolicyPath: vi.fn(),
+        getAutoSavedPolicyPath: vi.fn(),
+      } as unknown as Storage,
+
+      get config() {
+        return this;
+      },
+    } as unknown as Mocked<Config>;
+    mockSettings = {
+      merged: {
+        security: { auth: { selectedType: 'login_with_google' } },
+        mcpServers: {},
+      },
+      setValue: vi.fn(),
+    } as unknown as Mocked<LoadedSettings>;
+    mockArgv = {} as unknown as CliArgs;
+    mockConnection = {
+      sessionUpdate: vi.fn(),
+      requestPermission: vi.fn(),
+    } as unknown as Mocked<acp.AgentSideConnection>;
+
+    (loadCliConfig as unknown as Mock).mockResolvedValue(mockConfig);
+    (loadSettings as unknown as Mock).mockImplementation(() => ({
+      merged: {
+        security: {
+          auth: { selectedType: AuthType.LOGIN_WITH_GOOGLE },
+          enablePermanentToolApproval: true,
+        },
+        mcpServers: {},
+      },
+      setValue: vi.fn(),
+    }));
+
+    manager = new AcpSessionManager(mockSettings, mockArgv, mockConnection);
+    vi.mock('node:crypto', () => ({
+      randomUUID: () => 'test-session-id',
+    }));
+  });
+
+  afterEach(() => {
+    vi.restoreAllMocks();
+  });
+
+  it('should create a new session', async () => {
+    vi.useFakeTimers();
+    mockConfig.getContentGeneratorConfig = vi.fn().mockReturnValue({
+      apiKey: 'test-key',
+    });
+    const response = await manager.newSession(
+      {
+        cwd: '/tmp',
+        mcpServers: [],
+      },
+      {},
+    );
+
+    expect(response.sessionId).toBe('test-session-id');
+    expect(loadCliConfig).toHaveBeenCalled();
+    expect(mockConfig.initialize).toHaveBeenCalled();
+    expect(mockConfig.getGeminiClient).toHaveBeenCalled();
+
+    // Verify deferred call (sendAvailableCommands)
+    await vi.runAllTimersAsync();
+    expect(mockConnection.sessionUpdate).toHaveBeenCalledWith(
+      expect.objectContaining({
+        update: expect.objectContaining({
+          sessionUpdate: 'available_commands_update',
+        }),
+      }),
+    );
+    vi.useRealTimers();
+  });
+
+  it('should return modes without plan mode when plan is disabled', async () => {
+    mockConfig.getContentGeneratorConfig = vi.fn().mockReturnValue({
+      apiKey: 'test-key',
+    });
+    mockConfig.isPlanEnabled = vi.fn().mockReturnValue(false);
+    mockConfig.getApprovalMode = vi.fn().mockReturnValue('default');
+
+    const response = await manager.newSession(
+      {
+        cwd: '/tmp',
+        mcpServers: [],
+      },
+      {},
+    );
+
+    expect(response.modes).toEqual({
+      availableModes: [
+        { id: 'default', name: 'Default', description: 'Prompts for approval' },
+        {
+          id: 'autoEdit',
+          name: 'Auto Edit',
+          description: 'Auto-approves edit tools',
+        },
+        { id: 'yolo', name: 'YOLO', description: 'Auto-approves all tools' },
+      ],
+      currentModeId: 'default',
+    });
+  });
+
+  it('should include preview models when user has access', async () => {
+    mockConfig.getContentGeneratorConfig = vi.fn().mockReturnValue({
+      apiKey: 'test-key',
+    });
+    mockConfig.getHasAccessToPreviewModel = vi.fn().mockReturnValue(true);
+    mockConfig.getGemini31LaunchedSync = vi.fn().mockReturnValue(true);
+
+    const response = await manager.newSession(
+      {
+        cwd: '/tmp',
+        mcpServers: [],
+      },
+      {},
+    );
+
+    expect(response.models?.availableModels).toEqual(
+      expect.arrayContaining([
+        expect.objectContaining({
+          modelId: 'auto-gemini-3',
+          name: expect.stringContaining('Auto'),
+        }),
+      ]),
+    );
+  });
+
+  it('should include gemini-3.1-flash-lite when useGemini31FlashLite is true', async () => {
+    mockConfig.getContentGeneratorConfig = vi.fn().mockReturnValue({
+      apiKey: 'test-key',
+    });
+    mockConfig.getHasAccessToPreviewModel = vi.fn().mockReturnValue(true);
+    mockConfig.getGemini31LaunchedSync = vi.fn().mockReturnValue(true);
+    mockConfig.getGemini31FlashLiteLaunchedSync = vi.fn().mockReturnValue(true);
+
+    const response = await manager.newSession(
+      {
+        cwd: '/tmp',
+        mcpServers: [],
+      },
+      {},
+    );
+
+    expect(response.models?.availableModels).toEqual(
+      expect.arrayContaining([
+        expect.objectContaining({
+          modelId: 'gemini-3.1-flash-lite-preview',
+          name: 'gemini-3.1-flash-lite-preview',
+        }),
+      ]),
+    );
+  });
+
+  it('should return modes with plan mode when plan is enabled', async () => {
+    mockConfig.getContentGeneratorConfig = vi.fn().mockReturnValue({
+      apiKey: 'test-key',
+    });
+    mockConfig.isPlanEnabled = vi.fn().mockReturnValue(true);
+    mockConfig.getApprovalMode = vi.fn().mockReturnValue('plan');
+
+    const response = await manager.newSession(
+      {
+        cwd: '/tmp',
+        mcpServers: [],
+      },
+      {},
+    );
+
+    expect(response.modes).toEqual({
+      availableModes: [
+        { id: 'default', name: 'Default', description: 'Prompts for approval' },
+        {
+          id: 'autoEdit',
+          name: 'Auto Edit',
+          description: 'Auto-approves edit tools',
+        },
+        { id: 'yolo', name: 'YOLO', description: 'Auto-approves all tools' },
+        { id: 'plan', name: 'Plan', description: 'Read-only mode' },
+      ],
+      currentModeId: 'plan',
+    });
+  });
+
+  it('should fail session creation if Gemini API key is missing', async () => {
+    (loadSettings as unknown as Mock).mockImplementation(() => ({
+      merged: {
+        security: { auth: { selectedType: AuthType.USE_GEMINI } },
+        mcpServers: {},
+      },
+      setValue: vi.fn(),
+    }));
+    mockConfig.getContentGeneratorConfig = vi.fn().mockReturnValue({
+      apiKey: undefined,
+    });
+
+    await expect(
+      manager.newSession(
+        {
+          cwd: '/tmp',
+          mcpServers: [],
+        },
+        {},
+      ),
+    ).rejects.toMatchObject({
+      message: 'Gemini API key is missing or not configured.',
+    });
+  });
+
+  it('should create a new session with mcp servers', async () => {
+    mockConfig.getContentGeneratorConfig = vi.fn().mockReturnValue({
+      apiKey: 'test-key',
+    });
+    const mcpServers = [
+      {
+        name: 'test-server',
+        command: 'node',
+        args: ['server.js'],
+        env: [{ name: 'KEY', value: 'VALUE' }],
+      },
+    ];
+
+    await manager.newSession(
+      {
+        cwd: '/tmp',
+        mcpServers,
+      },
+      {},
+    );
+
+    expect(loadCliConfig).toHaveBeenCalledWith(
+      expect.objectContaining({
+        mcpServers: expect.objectContaining({
+          'test-server': expect.objectContaining({
+            command: 'node',
+            args: ['server.js'],
+            env: { KEY: 'VALUE' },
+          }),
+        }),
+      }),
+      'test-session-id',
+      mockArgv,
+      { cwd: '/tmp' },
+    );
+  });
+
+  it('should handle authentication failure gracefully', async () => {
+    mockConfig.refreshAuth.mockRejectedValue(new Error('Auth failed'));
+
+    await expect(
+      manager.newSession(
+        {
+          cwd: '/tmp',
+          mcpServers: [],
+        },
+        {},
+      ),
+    ).rejects.toMatchObject({
+      message: 'Auth failed',
+    });
+  });
+
+  it('should initialize file system service if client supports it', async () => {
+    mockConfig.getContentGeneratorConfig = vi.fn().mockReturnValue({
+      apiKey: 'test-key',
+    });
+    manager.setClientCapabilities({
+      fs: { readTextFile: true, writeTextFile: true },
+    });
+
+    await manager.newSession(
+      {
+        cwd: '/tmp',
+        mcpServers: [],
+      },
+      {},
+    );
+
+    expect(mockConfig.setFileSystemService).toHaveBeenCalled();
+  });
+
+  it('should start auto memory for new ACP sessions', async () => {
+    mockConfig.getContentGeneratorConfig = vi.fn().mockReturnValue({
+      apiKey: 'test-key',
+    });
+
+    await manager.newSession(
+      {
+        cwd: '/tmp',
+        mcpServers: [],
+      },
+      {},
+    );
+
+    expect(startAutoMemoryIfEnabledMock).toHaveBeenCalledWith(mockConfig);
+  });
+});

packages/cli/src/acp/acpSessionManager.ts

@@ -0,0 +1,322 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import {
+  type Config,
+  AuthType,
+  MCPServerConfig,
+  debugLogger,
+  startupProfiler,
+  convertSessionToClientHistory,
+  createPolicyUpdater,
+} from '@google/gemini-cli-core';
+import * as acp from '@agentclientprotocol/sdk';
+import { randomUUID } from 'node:crypto';
+import { loadSettings, type LoadedSettings } from '../config/settings.js';
+import { SessionSelector } from '../utils/sessionUtils.js';
+import { Session } from './acpSession.js';
+import { AcpFileSystemService } from './acpFileSystemService.js';
+import { getAcpErrorMessage } from './acpErrors.js';
+import { buildAvailableModels, buildAvailableModes } from './acpUtils.js';
+import { loadCliConfig, type CliArgs } from '../config/config.js';
+import { startAutoMemoryIfEnabled } from '../utils/autoMemory.js';
+
+export interface AuthDetails {
+  apiKey?: string;
+  baseUrl?: string;
+  customHeaders?: Record<string, string>;
+}
+
+export class AcpSessionManager {
+  private sessions: Map<string, Session> = new Map();
+  private clientCapabilities: acp.ClientCapabilities | undefined;
+
+  constructor(
+    private settings: LoadedSettings,
+    private argv: CliArgs,
+    private connection: acp.AgentSideConnection,
+  ) {}
+
+  setClientCapabilities(capabilities: acp.ClientCapabilities) {
+    this.clientCapabilities = capabilities;
+  }
+
+  getSession(sessionId: string): Session | undefined {
+    return this.sessions.get(sessionId);
+  }
+
+  async newSession(
+    { cwd, mcpServers }: acp.NewSessionRequest,
+    authDetails: AuthDetails,
+  ): Promise<acp.NewSessionResponse> {
+    const sessionId = randomUUID();
+    const loadedSettings = loadSettings(cwd);
+    const config = await this.newSessionConfig(
+      sessionId,
+      cwd,
+      mcpServers,
+      loadedSettings,
+    );
+
+    const authType =
+      loadedSettings.merged.security.auth.selectedType || AuthType.USE_GEMINI;
+
+    let isAuthenticated = false;
+    let authErrorMessage = '';
+    try {
+      await config.refreshAuth(
+        authType,
+        authDetails.apiKey,
+        authDetails.baseUrl,
+        authDetails.customHeaders,
+      );
+      isAuthenticated = true;
+
+      // Extra validation for Gemini API key
+      const contentGeneratorConfig = config.getContentGeneratorConfig();
+      if (
+        authType === AuthType.USE_GEMINI &&
+        (!contentGeneratorConfig || !contentGeneratorConfig.apiKey)
+      ) {
+        isAuthenticated = false;
+        authErrorMessage = 'Gemini API key is missing or not configured.';
+      }
+    } catch (e) {
+      isAuthenticated = false;
+      authErrorMessage = getAcpErrorMessage(e);
+      debugLogger.error(
+        `Authentication failed: ${e instanceof Error ? e.stack : e}`,
+      );
+    }
+
+    if (!isAuthenticated) {
+      throw new acp.RequestError(
+        -32000,
+        authErrorMessage || 'Authentication required.',
+      );
+    }
+
+    if (this.clientCapabilities?.fs) {
+      const acpFileSystemService = new AcpFileSystemService(
+        this.connection,
+        sessionId,
+        this.clientCapabilities.fs,
+        config.getFileSystemService(),
+        cwd,
+      );
+      config.setFileSystemService(acpFileSystemService);
+    }
+
+    await config.initialize();
+    startupProfiler.flush(config);
+    startAutoMemoryIfEnabled(config);
+
+    const geminiClient = config.getGeminiClient();
+
+    const chat = await geminiClient.startChat();
+
+    const session = new Session(
+      sessionId,
+      chat,
+      config,
+      this.connection,
+      this.settings,
+    );
+    this.sessions.set(sessionId, session);
+
+    setTimeout(() => {
+      // eslint-disable-next-line @typescript-eslint/no-floating-promises
+      session.sendAvailableCommands();
+    }, 0);
+
+    const { availableModels, currentModelId } = buildAvailableModels(
+      config,
+      loadedSettings,
+    );
+
+    const response = {
+      sessionId,
+      modes: {
+        availableModes: buildAvailableModes(config.isPlanEnabled()),
+        currentModeId: config.getApprovalMode(),
+      },
+      models: {
+        availableModels,
+        currentModelId,
+      },
+    };
+    return response;
+  }
+
+  async loadSession(
+    { sessionId, cwd, mcpServers }: acp.LoadSessionRequest,
+    authDetails: AuthDetails,
+  ): Promise<acp.LoadSessionResponse> {
+    const config = await this.initializeSessionConfig(
+      sessionId,
+      cwd,
+      mcpServers,
+      authDetails,
+    );
+
+    const sessionSelector = new SessionSelector(config.storage);
+
+    const { sessionData, sessionPath } =
+      await sessionSelector.resolveSession(sessionId);
+
+    const clientHistory = convertSessionToClientHistory(sessionData.messages);
+
+    const geminiClient = config.getGeminiClient();
+    await geminiClient.initialize();
+    await geminiClient.resumeChat(clientHistory, {
+      conversation: sessionData,
+      filePath: sessionPath,
+    });
+
+    const session = new Session(
+      sessionId,
+      geminiClient.getChat(),
+      config,
+      this.connection,
+      this.settings,
+    );
+    this.sessions.set(sessionId, session);
+
+    // Stream history back to client
+    // eslint-disable-next-line @typescript-eslint/no-floating-promises
+    session.streamHistory(sessionData.messages);
+
+    setTimeout(() => {
+      // eslint-disable-next-line @typescript-eslint/no-floating-promises
+      session.sendAvailableCommands();
+    }, 0);
+
+    const { availableModels, currentModelId } = buildAvailableModels(
+      config,
+      this.settings,
+    );
+
+    const response = {
+      modes: {
+        availableModes: buildAvailableModes(config.isPlanEnabled()),
+        currentModeId: config.getApprovalMode(),
+      },
+      models: {
+        availableModels,
+        currentModelId,
+      },
+    };
+    return response;
+  }
+
+  private async initializeSessionConfig(
+    sessionId: string,
+    cwd: string,
+    mcpServers: acp.McpServer[],
+    authDetails: AuthDetails,
+  ): Promise<Config> {
+    const selectedAuthType = this.settings.merged.security.auth.selectedType;
+    if (!selectedAuthType) {
+      throw acp.RequestError.authRequired();
+    }
+
+    // 1. Create config WITHOUT initializing it (no MCP servers started yet)
+    const config = await this.newSessionConfig(sessionId, cwd, mcpServers);
+
+    // 2. Authenticate BEFORE initializing configuration or starting MCP servers.
+    // This satisfies the security requirement to verify the user before executing
+    // potentially unsafe server definitions.
+    try {
+      await config.refreshAuth(
+        selectedAuthType,
+        authDetails.apiKey,
+        authDetails.baseUrl,
+        authDetails.customHeaders,
+      );
+    } catch (e) {
+      debugLogger.error(`Authentication failed: ${e}`);
+      throw acp.RequestError.authRequired();
+    }
+
+    // 3. Set the ACP FileSystemService (if supported) before config initialization
+    if (this.clientCapabilities?.fs) {
+      const acpFileSystemService = new AcpFileSystemService(
+        this.connection,
+        sessionId,
+        this.clientCapabilities.fs,
+        config.getFileSystemService(),
+        cwd,
+      );
+      config.setFileSystemService(acpFileSystemService);
+    }
+
+    // 4. Now that we are authenticated, it is safe to initialize the config
+    // which starts the MCP servers and other heavy resources.
+    await config.initialize();
+    startupProfiler.flush(config);
+    startAutoMemoryIfEnabled(config);
+
+    return config;
+  }
+
+  async newSessionConfig(
+    sessionId: string,
+    cwd: string,
+    mcpServers: acp.McpServer[],
+    loadedSettings?: LoadedSettings,
+  ): Promise<Config> {
+    const currentSettings = loadedSettings || this.settings;
+    const mergedMcpServers = { ...currentSettings.merged.mcpServers };
+
+    for (const server of mcpServers) {
+      if (
+        'type' in server &&
+        (server.type === 'sse' || server.type === 'http')
+      ) {
+        // HTTP or SSE MCP server
+        const headers = Object.fromEntries(
+          server.headers.map(({ name, value }) => [name, value]),
+        );
+        mergedMcpServers[server.name] = new MCPServerConfig(
+          undefined, // command
+          undefined, // args
+          undefined, // env
+          undefined, // cwd
+          server.type === 'sse' ? server.url : undefined, // url (sse)
+          server.type === 'http' ? server.url : undefined, // httpUrl
+          headers,
+        );
+      } else if ('command' in server) {
+        // Stdio MCP server
+        const env: Record<string, string> = {};
+        for (const { name: envName, value } of server.env) {
+          env[envName] = value;
+        }
+        mergedMcpServers[server.name] = new MCPServerConfig(
+          server.command,
+          server.args,
+          env,
+          cwd,
+        );
+      }
+    }
+
+    const settings = {
+      ...currentSettings.merged,
+      mcpServers: mergedMcpServers,
+    };
+
+    const config = await loadCliConfig(settings, sessionId, this.argv, { cwd });
+
+    createPolicyUpdater(
+      config.getPolicyEngine(),
+      config.messageBus,
+      config.storage,
+    );
+
+    return config;
+  }
+}

packages/cli/src/acp/acpStdioTransport.ts

@@ -0,0 +1,35 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import { type Config, createWorkingStdio } from '@google/gemini-cli-core';
+import { runExitCleanup } from '../utils/cleanup.js';
+import * as acp from '@agentclientprotocol/sdk';
+import { Readable, Writable } from 'node:stream';
+import type { LoadedSettings } from '../config/settings.js';
+import type { CliArgs } from '../config/config.js';
+import { GeminiAgent } from './acpRpcDispatcher.js';
+
+export async function runAcpClient(
+  config: Config,
+  settings: LoadedSettings,
+  argv: CliArgs,
+) {
+  const { stdout: workingStdout } = createWorkingStdio();
+  const stdout = Writable.toWeb(workingStdout) as WritableStream;
+  // eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
+  const stdin = Readable.toWeb(process.stdin) as ReadableStream<Uint8Array>;
+
+  const stream = acp.ndJsonStream(stdout, stdin);
+  const connection = new acp.AgentSideConnection(
+    (connection) => new GeminiAgent(config, settings, argv, connection),
+    stream,
+  );
+
+  // SIGTERM/SIGINT handlers (in sdk.ts) don't fire when stdin closes.
+  // We must explicitly await the connection close to flush telemetry.
+  // Use finally() to ensure cleanup runs even on stream errors.
+  await connection.closed.finally(runExitCleanup);
+}

packages/cli/src/acp/acpUtils.ts

@@ -0,0 +1,373 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import {
+  type Config,
+  type ToolResult,
+  type ToolCallConfirmationDetails,
+  Kind,
+  ApprovalMode,
+  DEFAULT_GEMINI_MODEL_AUTO,
+  PREVIEW_GEMINI_MODEL_AUTO,
+  DEFAULT_GEMINI_MODEL,
+  DEFAULT_GEMINI_FLASH_MODEL,
+  DEFAULT_GEMINI_FLASH_LITE_MODEL,
+  PREVIEW_GEMINI_3_1_MODEL,
+  PREVIEW_GEMINI_MODEL,
+  PREVIEW_GEMINI_3_1_CUSTOM_TOOLS_MODEL,
+  PREVIEW_GEMINI_FLASH_MODEL,
+  PREVIEW_GEMINI_3_1_FLASH_LITE_MODEL,
+  getDisplayString,
+  AuthType,
+  ToolConfirmationOutcome,
+} from '@google/gemini-cli-core';
+import type * as acp from '@agentclientprotocol/sdk';
+import { z } from 'zod';
+import type { LoadedSettings } from '../config/settings.js';
+
+export function hasMeta(
+  obj: unknown,
+): obj is { _meta?: Record<string, unknown> } {
+  return typeof obj === 'object' && obj !== null && '_meta' in obj;
+}
+
+export const RequestPermissionResponseSchema = z.object({
+  outcome: z.discriminatedUnion('outcome', [
+    z.object({ outcome: z.literal('cancelled') }),
+    z.object({
+      outcome: z.literal('selected'),
+      optionId: z.string(),
+    }),
+  ]),
+});
+
+export function toToolCallContent(
+  toolResult: ToolResult,
+): acp.ToolCallContent | null {
+  if (toolResult.error?.message) {
+    throw new Error(toolResult.error.message);
+  }
+
+  if (toolResult.returnDisplay) {
+    if (typeof toolResult.returnDisplay === 'string') {
+      return {
+        type: 'content',
+        content: { type: 'text', text: toolResult.returnDisplay },
+      };
+    } else {
+      if ('fileName' in toolResult.returnDisplay) {
+        return {
+          type: 'diff',
+          path:
+            toolResult.returnDisplay.filePath ??
+            toolResult.returnDisplay.fileName,
+          oldText: toolResult.returnDisplay.originalContent,
+          newText: toolResult.returnDisplay.newContent,
+          _meta: {
+            kind: !toolResult.returnDisplay.originalContent
+              ? 'add'
+              : toolResult.returnDisplay.newContent === ''
+                ? 'delete'
+                : 'modify',
+          },
+        };
+      }
+      return null;
+    }
+  } else {
+    return null;
+  }
+}
+
+const basicPermissionOptions = [
+  {
+    optionId: ToolConfirmationOutcome.ProceedOnce,
+    name: 'Allow',
+    kind: 'allow_once',
+  },
+  {
+    optionId: ToolConfirmationOutcome.Cancel,
+    name: 'Reject',
+    kind: 'reject_once',
+  },
+] as const;
+
+export function toPermissionOptions(
+  confirmation: ToolCallConfirmationDetails,
+  config: Config,
+  enablePermanentToolApproval: boolean = false,
+): acp.PermissionOption[] {
+  const disableAlwaysAllow = config.getDisableAlwaysAllow();
+  const options: acp.PermissionOption[] = [];
+
+  if (!disableAlwaysAllow) {
+    switch (confirmation.type) {
+      case 'edit':
+        options.push({
+          optionId: ToolConfirmationOutcome.ProceedAlways,
+          name: 'Allow for this session',
+          kind: 'allow_always',
+        });
+        if (enablePermanentToolApproval) {
+          options.push({
+            optionId: ToolConfirmationOutcome.ProceedAlwaysAndSave,
+            name: 'Allow for this file in all future sessions',
+            kind: 'allow_always',
+          });
+        }
+        break;
+      case 'exec':
+        options.push({
+          optionId: ToolConfirmationOutcome.ProceedAlways,
+          name: 'Allow for this session',
+          kind: 'allow_always',
+        });
+        if (enablePermanentToolApproval) {
+          options.push({
+            optionId: ToolConfirmationOutcome.ProceedAlwaysAndSave,
+            name: 'Allow this command for all future sessions',
+            kind: 'allow_always',
+          });
+        }
+        break;
+      case 'mcp':
+        options.push(
+          {
+            optionId: ToolConfirmationOutcome.ProceedAlwaysServer,
+            name: 'Allow all server tools for this session',
+            kind: 'allow_always',
+          },
+          {
+            optionId: ToolConfirmationOutcome.ProceedAlwaysTool,
+            name: 'Allow tool for this session',
+            kind: 'allow_always',
+          },
+        );
+        if (enablePermanentToolApproval) {
+          options.push({
+            optionId: ToolConfirmationOutcome.ProceedAlwaysAndSave,
+            name: 'Allow tool for all future sessions',
+            kind: 'allow_always',
+          });
+        }
+        break;
+      case 'info':
+        options.push({
+          optionId: ToolConfirmationOutcome.ProceedAlways,
+          name: 'Allow for this session',
+          kind: 'allow_always',
+        });
+        if (enablePermanentToolApproval) {
+          options.push({
+            optionId: ToolConfirmationOutcome.ProceedAlwaysAndSave,
+            name: 'Allow for all future sessions',
+            kind: 'allow_always',
+          });
+        }
+        break;
+      case 'ask_user':
+      case 'exit_plan_mode':
+        // askuser and exit_plan_mode don't need "always allow" options
+        break;
+      default:
+        // No "always allow" options for other types
+        break;
+    }
+  }
+
+  options.push(...basicPermissionOptions);
+
+  // Exhaustive check
+  switch (confirmation.type) {
+    case 'edit':
+    case 'exec':
+    case 'mcp':
+    case 'info':
+    case 'ask_user':
+    case 'exit_plan_mode':
+    case 'sandbox_expansion':
+      break;
+    default: {
+      const unreachable: never = confirmation;
+      throw new Error(`Unexpected: ${unreachable}`);
+    }
+  }
+
+  return options;
+}
+
+export function toAcpToolKind(kind: Kind): acp.ToolKind {
+  switch (kind) {
+    case Kind.Read:
+    case Kind.Edit:
+    case Kind.Execute:
+    case Kind.Search:
+    case Kind.Delete:
+    case Kind.Move:
+    case Kind.Think:
+    case Kind.Fetch:
+    case Kind.SwitchMode:
+    case Kind.Other:
+      return kind as acp.ToolKind;
+    case Kind.Agent:
+      return 'think';
+    case Kind.Plan:
+    case Kind.Communicate:
+    default:
+      return 'other';
+  }
+}
+
+export function buildAvailableModes(isPlanEnabled: boolean): acp.SessionMode[] {
+  const modes: acp.SessionMode[] = [
+    {
+      id: ApprovalMode.DEFAULT,
+      name: 'Default',
+      description: 'Prompts for approval',
+    },
+    {
+      id: ApprovalMode.AUTO_EDIT,
+      name: 'Auto Edit',
+      description: 'Auto-approves edit tools',
+    },
+    {
+      id: ApprovalMode.YOLO,
+      name: 'YOLO',
+      description: 'Auto-approves all tools',
+    },
+  ];
+
+  if (isPlanEnabled) {
+    modes.push({
+      id: ApprovalMode.PLAN,
+      name: 'Plan',
+      description: 'Read-only mode',
+    });
+  }
+
+  return modes;
+}
+
+export function buildAvailableModels(
+  config: Config,
+  settings: LoadedSettings,
+): {
+  availableModels: Array<{
+    modelId: string;
+    name: string;
+    description?: string;
+  }>;
+  currentModelId: string;
+} {
+  const preferredModel = config.getModel() || DEFAULT_GEMINI_MODEL_AUTO;
+  const shouldShowPreviewModels = config.getHasAccessToPreviewModel();
+  const useGemini31 = config.getGemini31LaunchedSync?.() ?? false;
+  const useGemini31FlashLite =
+    config.getGemini31FlashLiteLaunchedSync?.() ?? false;
+  const selectedAuthType = settings.merged.security.auth.selectedType;
+  const useCustomToolModel =
+    useGemini31 && selectedAuthType === AuthType.USE_GEMINI;
+
+  // --- DYNAMIC PATH ---
+  if (
+    config.getExperimentalDynamicModelConfiguration?.() === true &&
+    config.getModelConfigService
+  ) {
+    const options = config.getModelConfigService().getAvailableModelOptions({
+      useGemini3_1: useGemini31,
+      useGemini3_1FlashLite: useGemini31FlashLite,
+      useCustomTools: useCustomToolModel,
+      hasAccessToPreview: shouldShowPreviewModels,
+    });
+
+    return {
+      availableModels: options,
+      currentModelId: preferredModel,
+    };
+  }
+
+  // --- LEGACY PATH ---
+  const mainOptions = [
+    {
+      value: DEFAULT_GEMINI_MODEL_AUTO,
+      title: getDisplayString(DEFAULT_GEMINI_MODEL_AUTO),
+      description:
+        'Let Gemini CLI decide the best model for the task: gemini-2.5-pro, gemini-2.5-flash',
+    },
+  ];
+
+  if (shouldShowPreviewModels) {
+    mainOptions.unshift({
+      value: PREVIEW_GEMINI_MODEL_AUTO,
+      title: getDisplayString(PREVIEW_GEMINI_MODEL_AUTO),
+      description: useGemini31
+        ? 'Let Gemini CLI decide the best model for the task: gemini-3.1-pro, gemini-3-flash'
+        : 'Let Gemini CLI decide the best model for the task: gemini-3-pro, gemini-3-flash',
+    });
+  }
+
+  const manualOptions = [
+    {
+      value: DEFAULT_GEMINI_MODEL,
+      title: getDisplayString(DEFAULT_GEMINI_MODEL),
+    },
+    {
+      value: DEFAULT_GEMINI_FLASH_MODEL,
+      title: getDisplayString(DEFAULT_GEMINI_FLASH_MODEL),
+    },
+    {
+      value: DEFAULT_GEMINI_FLASH_LITE_MODEL,
+      title: getDisplayString(DEFAULT_GEMINI_FLASH_LITE_MODEL),
+    },
+  ];
+
+  if (shouldShowPreviewModels) {
+    const previewProModel = useGemini31
+      ? PREVIEW_GEMINI_3_1_MODEL
+      : PREVIEW_GEMINI_MODEL;
+
+    const previewProValue = useCustomToolModel
+      ? PREVIEW_GEMINI_3_1_CUSTOM_TOOLS_MODEL
+      : previewProModel;
+
+    const previewOptions = [
+      {
+        value: previewProValue,
+        title: getDisplayString(previewProModel),
+      },
+      {
+        value: PREVIEW_GEMINI_FLASH_MODEL,
+        title: getDisplayString(PREVIEW_GEMINI_FLASH_MODEL),
+      },
+    ];
+
+    if (useGemini31FlashLite) {
+      previewOptions.push({
+        value: PREVIEW_GEMINI_3_1_FLASH_LITE_MODEL,
+        title: getDisplayString(PREVIEW_GEMINI_3_1_FLASH_LITE_MODEL),
+      });
+    }
+
+    manualOptions.unshift(...previewOptions);
+  }
+
+  const scaleOptions = (
+    options: Array<{ value: string; title: string; description?: string }>,
+  ) =>
+    options.map((o) => ({
+      modelId: o.value,
+      name: o.title,
+      description: o.description,
+    }));
+
+  return {
+    availableModels: [
+      ...scaleOptions(mainOptions),
+      ...scaleOptions(manualOptions),
+    ],
+    currentModelId: preferredModel,
+  };
+}

packages/cli/src/acp/commands/commandRegistry.ts

@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2025 Google LLC
+ * Copyright 2026 Google LLC
  * SPDX-License-Identifier: Apache-2.0
  */
 

packages/cli/src/acp/commands/extensions.ts

@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2025 Google LLC
+ * Copyright 2026 Google LLC
  * SPDX-License-Identifier: Apache-2.0
  */
 

packages/cli/src/acp/commands/init.ts

@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2025 Google LLC
+ * Copyright 2026 Google LLC
  * SPDX-License-Identifier: Apache-2.0
  */
 

packages/cli/src/acp/commands/memory.ts

@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2025 Google LLC
+ * Copyright 2026 Google LLC
  * SPDX-License-Identifier: Apache-2.0
  */
 

packages/cli/src/acp/commands/restore.test.ts

@@ -0,0 +1,244 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import { describe, it, expect, vi, beforeEach } from 'vitest';
+import { RestoreCommand, ListCheckpointsCommand } from './restore.js';
+import * as fs from 'node:fs/promises';
+import {
+  getCheckpointInfoList,
+  getToolCallDataSchema,
+  isNodeError,
+  performRestore,
+} from '@google/gemini-cli-core';
+import type { CommandContext } from './types.js';
+import type { Mock } from 'vitest';
+
+vi.mock('node:fs/promises');
+vi.mock('@google/gemini-cli-core', async (importOriginal) => {
+  const actual =
+    await importOriginal<typeof import('@google/gemini-cli-core')>();
+  return {
+    ...actual,
+    getCheckpointInfoList: vi.fn(),
+    getToolCallDataSchema: vi.fn(),
+    isNodeError: vi.fn(),
+    performRestore: vi.fn(),
+  };
+});
+
+describe('RestoreCommand', () => {
+  let context: CommandContext;
+  let restoreCommand: RestoreCommand;
+
+  beforeEach(() => {
+    vi.resetAllMocks();
+    restoreCommand = new RestoreCommand();
+    context = {
+      agentContext: {
+        config: {
+          getCheckpointingEnabled: vi.fn().mockReturnValue(true),
+          storage: {
+            getProjectTempCheckpointsDir: vi
+              .fn()
+              .mockReturnValue('/tmp/checkpoints'),
+          },
+        },
+      },
+      git: {},
+      sendMessage: vi.fn(),
+    } as unknown as CommandContext;
+  });
+
+  it('delegates to list behavior when invoked without args', async () => {
+    const listExecuteSpy = vi
+      .spyOn(ListCheckpointsCommand.prototype, 'execute')
+      .mockResolvedValue({
+        name: 'restore list',
+        data: 'list data',
+      });
+
+    const response = await restoreCommand.execute(context, []);
+
+    expect(listExecuteSpy).toHaveBeenCalledWith(context);
+    expect(response).toEqual({
+      name: 'restore list',
+      data: 'list data',
+    });
+  });
+
+  it('returns checkpointing-disabled message when disabled', async () => {
+    (
+      context.agentContext.config.getCheckpointingEnabled as Mock
+    ).mockReturnValue(false);
+
+    const response = await restoreCommand.execute(context, ['checkpoint1']);
+
+    expect(response.data).toContain('Checkpointing is not enabled');
+  });
+
+  it('returns file-not-found message for missing checkpoint', async () => {
+    const error = new Error('ENOENT');
+    (error as Error & { code: string }).code = 'ENOENT';
+    vi.mocked(fs.readFile).mockRejectedValue(error);
+    vi.mocked(isNodeError).mockReturnValue(true);
+
+    const response = await restoreCommand.execute(context, ['missing']);
+
+    expect(response.data).toBe('File not found: missing.json');
+  });
+
+  it('handles checkpoint filename already ending in .json', async () => {
+    const error = new Error('ENOENT');
+    (error as Error & { code: string }).code = 'ENOENT';
+    vi.mocked(fs.readFile).mockRejectedValue(error);
+    vi.mocked(isNodeError).mockReturnValue(true);
+
+    const response = await restoreCommand.execute(context, ['existing.json']);
+
+    expect(response.data).toBe('File not found: existing.json');
+    expect(fs.readFile).toHaveBeenCalledWith(
+      expect.stringContaining('existing.json'),
+      'utf-8',
+    );
+  });
+
+  it('returns invalid/corrupt checkpoint message when schema parse fails', async () => {
+    vi.mocked(fs.readFile).mockResolvedValue('{"invalid": "data"}');
+    vi.mocked(getToolCallDataSchema).mockReturnValue({
+      safeParse: vi.fn().mockReturnValue({ success: false }),
+    } as unknown as ReturnType<typeof getToolCallDataSchema>);
+
+    const response = await restoreCommand.execute(context, ['invalid']);
+
+    expect(response.data).toBe('Checkpoint file is invalid or corrupted.');
+  });
+
+  it('formats streamed restore results correctly', async () => {
+    vi.mocked(fs.readFile).mockResolvedValue('{"valid": "data"}');
+    vi.mocked(getToolCallDataSchema).mockReturnValue({
+      safeParse: vi
+        .fn()
+        .mockReturnValue({ success: true, data: { some: 'data' } }),
+    } as unknown as ReturnType<typeof getToolCallDataSchema>);
+
+    async function* mockRestoreGenerator() {
+      yield { type: 'message', messageType: 'info', content: 'Restoring...' };
+      yield { type: 'load_history', clientHistory: [{}, {}] };
+      yield { type: 'other', some: 'other' };
+    }
+    vi.mocked(performRestore).mockReturnValue(
+      mockRestoreGenerator() as unknown as ReturnType<typeof performRestore>,
+    );
+
+    const response = await restoreCommand.execute(context, ['valid']);
+
+    expect(response.data).toContain('[INFO] Restoring...');
+    expect(response.data).toContain('Loaded history with 2 messages.');
+    expect(response.data).toContain(
+      'Restored: {"type":"other","some":"other"}',
+    );
+  });
+
+  it('returns generic unexpected error message for non-ENOENT failures', async () => {
+    vi.mocked(fs.readFile).mockRejectedValue(new Error('Random error'));
+    vi.mocked(isNodeError).mockReturnValue(false);
+
+    const response = await restoreCommand.execute(context, ['error']);
+
+    expect(response.data).toContain(
+      'An unexpected error occurred during restore: Error: Random error',
+    );
+  });
+});
+
+describe('ListCheckpointsCommand', () => {
+  let context: CommandContext;
+  let listCommand: ListCheckpointsCommand;
+  let mockReaddir: Mock<(path: string) => Promise<string[]>>;
+
+  beforeEach(() => {
+    vi.resetAllMocks();
+    listCommand = new ListCheckpointsCommand();
+    mockReaddir = vi.mocked(fs.readdir) as unknown as Mock<
+      (path: string) => Promise<string[]>
+    >;
+
+    context = {
+      agentContext: {
+        config: {
+          getCheckpointingEnabled: vi.fn().mockReturnValue(true),
+          storage: {
+            getProjectTempCheckpointsDir: vi
+              .fn()
+              .mockReturnValue('/tmp/checkpoints'),
+          },
+        },
+      },
+    } as unknown as CommandContext;
+  });
+
+  it('returns checkpointing-disabled message when disabled', async () => {
+    (
+      context.agentContext.config.getCheckpointingEnabled as Mock
+    ).mockReturnValue(false);
+
+    const response = await listCommand.execute(context);
+
+    expect(response.data).toContain('Checkpointing is not enabled');
+  });
+
+  it('returns "No checkpoints found." when no .json checkpoints exist', async () => {
+    mockReaddir.mockResolvedValue(['not-a-checkpoint.txt']);
+
+    const response = await listCommand.execute(context);
+
+    expect(response.data).toBe('No checkpoints found.');
+  });
+
+  it('ignores error when mkdir fails', async () => {
+    vi.mocked(fs.mkdir).mockRejectedValue(new Error('mkdir fail'));
+    mockReaddir.mockResolvedValue([]);
+
+    const response = await listCommand.execute(context);
+
+    expect(response.data).toBe('No checkpoints found.');
+    expect(fs.mkdir).toHaveBeenCalled();
+  });
+
+  it('formats checkpoint summary output from checkpoint metadata', async () => {
+    mockReaddir.mockResolvedValue(['cp1.json', 'cp2.json']);
+    vi.mocked(getCheckpointInfoList).mockReturnValue([
+      { messageId: 'id1', checkpoint: 'cp1' },
+      { messageId: 'id2', checkpoint: 'cp2' },
+    ]);
+
+    const response = await listCommand.execute(context);
+
+    expect(response.data).toContain('Available Checkpoints:');
+    // Note: The current implementation of ListCheckpointsCommand incorrectly accesses
+    // fileName, toolName, etc. which don't exist on CheckpointInfo, resulting in 'Unknown'.
+    expect(response.data).toContain('- **Unknown**: Unknown (Status: Unknown)');
+  });
+
+  it('handles empty checkpoint info list', async () => {
+    mockReaddir.mockResolvedValue(['some.json']);
+    vi.mocked(getCheckpointInfoList).mockReturnValue([]);
+
+    const response = await listCommand.execute(context);
+
+    expect(response.data).toBe('Available Checkpoints:\n');
+  });
+
+  it('returns generic unexpected error message on failures', async () => {
+    mockReaddir.mockRejectedValue(new Error('Readdir fail'));
+
+    const response = await listCommand.execute(context);
+
+    expect(response.data).toBe(
+      'An unexpected error occurred while listing checkpoints.',
+    );
+  });
+});

packages/cli/src/acp/commands/restore.ts

@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2025 Google LLC
+ * Copyright 2026 Google LLC
  * SPDX-License-Identifier: Apache-2.0
  */
 

packages/cli/src/acp/commands/types.ts

@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2025 Google LLC
+ * Copyright 2026 Google LLC
  * SPDX-License-Identifier: Apache-2.0
  */
 

packages/cli/src/commands/mcp/list.test.ts

@@ -217,6 +217,78 @@ describe('mcp list command', () => {
     );
   });
 
+  it('should display connected status even if ping fails', async () => {
+    const defaultMergedSettings = mergeSettings({}, {}, {}, {}, true);
+    mockedLoadSettings.mockReturnValue({
+      merged: {
+        ...defaultMergedSettings,
+        mcpServers: {
+          'test-server': { command: '/test/server' },
+        },
+      },
+      isTrusted: true,
+    });
+
+    mockClient.connect.mockResolvedValue(undefined);
+    mockClient.ping.mockRejectedValue(new Error('Ping failed'));
+
+    await listMcpServers();
+
+    expect(debugLogger.log).toHaveBeenCalledWith(
+      expect.stringContaining('test-server: /test/server  (stdio) - Connected'),
+    );
+  });
+
+  it('should use configured timeout for connection', async () => {
+    const defaultMergedSettings = mergeSettings({}, {}, {}, {}, true);
+    mockedLoadSettings.mockReturnValue({
+      merged: {
+        ...defaultMergedSettings,
+        mcpServers: {
+          'test-server': { command: '/test/server', timeout: 12345 },
+        },
+      },
+      isTrusted: true,
+    });
+
+    mockClient.connect.mockResolvedValue(undefined);
+
+    await listMcpServers();
+
+    expect(mockClient.connect).toHaveBeenCalledWith(
+      expect.anything(),
+      expect.objectContaining({ timeout: 12345 }),
+    );
+    expect(mockClient.ping).toHaveBeenCalledWith(
+      expect.objectContaining({ timeout: 12345 }),
+    );
+  });
+
+  it('should use default timeout for connection when not configured', async () => {
+    const defaultMergedSettings = mergeSettings({}, {}, {}, {}, true);
+    mockedLoadSettings.mockReturnValue({
+      merged: {
+        ...defaultMergedSettings,
+        mcpServers: {
+          'test-server': { command: '/test/server' },
+        },
+      },
+      isTrusted: true,
+    });
+
+    mockClient.connect.mockResolvedValue(undefined);
+
+    await listMcpServers();
+
+    expect(mockClient.connect).toHaveBeenCalledWith(
+      expect.anything(),
+      expect.objectContaining({ timeout: 5000 }),
+    );
+    expect(mockClient.ping).toHaveBeenCalledWith(
+      expect.objectContaining({ timeout: 5000 }),
+    );
+  });
+
   it('should merge extension servers with config servers', async () => {
     const defaultMergedSettings = mergeSettings({}, {}, {}, {}, true);
     mockedLoadSettings.mockReturnValue({

packages/cli/src/commands/mcp/list.ts

@@ -67,6 +67,8 @@ export async function getMcpServersFromConfig(
   return filteredResult;
 }
 
+const MCP_LIST_DEFAULT_TIMEOUT_MSEC = 5000;
+
 async function testMCPConnection(
   serverName: string,
   config: MCPServerConfig,
@@ -127,11 +129,22 @@ async function testMCPConnection(
   }
 
   try {
-    // Attempt actual MCP connection with short timeout
-    await client.connect(transport, { timeout: 5000 }); // 5s timeout
+    // Attempt actual MCP connection with timeout from config or default to 5s.
+    // We use a short default for the list command to keep it responsive.
+    const timeout = config.timeout ?? MCP_LIST_DEFAULT_TIMEOUT_MSEC;
+    await client.connect(transport, { timeout });
 
-    // Test basic MCP protocol by pinging the server
-    await client.ping();
+    // Test basic MCP protocol by pinging the server.
+    // Ping is optional per MCP spec - some servers (e.g. Google first-party)
+    // don't implement it. A successful connect() is sufficient proof of connectivity.
+    try {
+      await client.ping({ timeout });
+    } catch (e) {
+      debugLogger.debug(
+        `MCP ping failed for ${serverName}, but connect succeeded:`,
+        e,
+      );
+    }
 
     await client.close();
     return MCPServerStatus.CONNECTED;

packages/cli/src/config/config.test.ts

@@ -21,8 +21,6 @@ import {
   type MCPServerConfig,
   type GeminiCLIExtension,
   Storage,
-  generalistProfile,
-  type ContextManagementConfig,
 } from '@google/gemini-cli-core';
 import { loadCliConfig, parseArguments, type CliArgs } from './config.js';
 import {
@@ -233,6 +231,45 @@ afterEach(() => {
 });
 
 describe('parseArguments', () => {
+  afterEach(() => {
+    vi.restoreAllMocks();
+  });
+  it('should fail if both --resume and --session-id are provided', async () => {
+    process.argv = [
+      'node',
+      'script.js',
+      '--resume',
+      '--session-id',
+      'test-uuid-1234',
+    ];
+    const mockConsoleError = vi
+      .spyOn(console, 'error')
+      .mockImplementation(() => {});
+    vi.spyOn(process, 'exit').mockImplementation(() => {
+      throw new Error('process.exit called');
+    });
+
+    await expect(parseArguments(createTestMergedSettings())).rejects.toThrow(
+      'process.exit called',
+    );
+
+    expect(mockConsoleError).toHaveBeenCalledWith(
+      expect.stringContaining(
+        'Cannot use both --resume (-r) and --session-id together',
+      ),
+    );
+  });
+
+  it('should parse --session-id option correctly', async () => {
+    process.argv = ['node', 'script.js', '--session-id', 'test-uuid-1234'];
+    vi.spyOn(process, 'exit').mockImplementation(() => {
+      throw new Error('process.exit called');
+    });
+
+    const parsedArgs = await parseArguments(createTestMergedSettings());
+    expect(parsedArgs.sessionId).toBe('test-uuid-1234');
+  });
+
   describe('worktree', () => {
     it('should parse --worktree flag when provided with a name', async () => {
       process.argv = ['node', 'script.js', '--worktree', 'my-feature'];
@@ -257,7 +294,7 @@ describe('parseArguments', () => {
       const settings = createTestMergedSettings();
       settings.experimental.worktrees = false;
 
-      const mockExit = vi.spyOn(process, 'exit').mockImplementation(() => {
+      vi.spyOn(process, 'exit').mockImplementation(() => {
         throw new Error('process.exit called');
       });
       const mockConsoleError = vi
@@ -272,9 +309,6 @@ describe('parseArguments', () => {
           'The --worktree flag is only available when experimental.worktrees is enabled in your settings.',
         ),
       );
-
-      mockExit.mockRestore();
-      mockConsoleError.mockRestore();
     });
   });
 
@@ -306,7 +340,7 @@ describe('parseArguments', () => {
     async ({ argv }) => {
       process.argv = argv;
 
-      const mockExit = vi.spyOn(process, 'exit').mockImplementation(() => {
+      vi.spyOn(process, 'exit').mockImplementation(() => {
         throw new Error('process.exit called');
       });
 
@@ -323,9 +357,6 @@ describe('parseArguments', () => {
           'Cannot use both --prompt (-p) and --prompt-interactive (-i) together',
         ),
       );
-
-      mockExit.mockRestore();
-      mockConsoleError.mockRestore();
     },
   );
 
@@ -562,7 +593,7 @@ describe('parseArguments', () => {
     async ({ argv }) => {
       process.argv = argv;
 
-      const mockExit = vi.spyOn(process, 'exit').mockImplementation(() => {
+      vi.spyOn(process, 'exit').mockImplementation(() => {
         throw new Error('process.exit called');
       });
 
@@ -579,9 +610,6 @@ describe('parseArguments', () => {
           'Cannot use both --yolo (-y) and --approval-mode together. Use --approval-mode=yolo instead.',
         ),
       );
-
-      mockExit.mockRestore();
-      mockConsoleError.mockRestore();
     },
   );
 
@@ -606,7 +634,7 @@ describe('parseArguments', () => {
   it('should reject invalid --approval-mode values', async () => {
     process.argv = ['node', 'script.js', '--approval-mode', 'invalid'];
 
-    const mockExit = vi.spyOn(process, 'exit').mockImplementation(() => {
+    vi.spyOn(process, 'exit').mockImplementation(() => {
       throw new Error('process.exit called');
     });
 
@@ -625,10 +653,6 @@ describe('parseArguments', () => {
       expect.stringContaining('Invalid values:'),
     );
     expect(mockConsoleError).toHaveBeenCalled();
-
-    mockExit.mockRestore();
-    mockConsoleError.mockRestore();
-    debugErrorSpy.mockRestore();
   });
 
   it('should allow resuming a session without prompt argument in non-interactive mode (expecting stdin)', async () => {
@@ -780,6 +804,100 @@ describe('loadCliConfig', () => {
     vi.restoreAllMocks();
   });
 
+  describe('Model resolution', () => {
+    it('should handle multiple --model flags by taking the last one', async () => {
+      const argv = {
+        query: undefined,
+        // eslint-disable-next-line @typescript-eslint/no-explicit-any
+        model: ['gemini-1.5-pro', 'gemini-2.0-flash'] as any,
+        sandbox: undefined,
+        debug: false,
+        prompt: undefined,
+        promptInteractive: undefined,
+        yolo: undefined,
+        approvalMode: undefined,
+        policy: undefined,
+        adminPolicy: undefined,
+        allowedMcpServerNames: undefined,
+        allowedTools: undefined,
+        extensions: undefined,
+        listExtensions: false,
+        listSessions: false,
+        deleteSession: undefined,
+        screenReader: undefined,
+        isCommand: false,
+        rawOutput: false,
+        acceptRawOutputRisk: false,
+        startupMessages: [],
+        resume: undefined,
+        includeDirectories: [],
+        useWriteTodos: false,
+        outputFormat: undefined,
+        fakeResponses: undefined,
+        recordResponses: undefined,
+        skipTrust: false,
+      };
+
+      const settings = createTestMergedSettings();
+      const config = await loadCliConfig(
+        settings,
+        'test-session',
+        argv as unknown as CliArgs,
+        {
+          cwd: process.cwd(),
+        },
+      );
+
+      expect(config.getModel()).toBe('gemini-2.0-flash');
+    });
+
+    it('should handle non-string model flags by coercing to string', async () => {
+      const argv = {
+        query: undefined,
+        // eslint-disable-next-line @typescript-eslint/no-explicit-any
+        model: true as any,
+        sandbox: undefined,
+        debug: false,
+        prompt: undefined,
+        promptInteractive: undefined,
+        yolo: undefined,
+        approvalMode: undefined,
+        policy: undefined,
+        adminPolicy: undefined,
+        allowedMcpServerNames: undefined,
+        allowedTools: undefined,
+        extensions: undefined,
+        listExtensions: false,
+        listSessions: false,
+        deleteSession: undefined,
+        screenReader: undefined,
+        isCommand: false,
+        rawOutput: false,
+        acceptRawOutputRisk: false,
+        startupMessages: [],
+        resume: undefined,
+        includeDirectories: [],
+        useWriteTodos: false,
+        outputFormat: undefined,
+        fakeResponses: undefined,
+        recordResponses: undefined,
+        skipTrust: false,
+      };
+
+      const settings = createTestMergedSettings();
+      const config = await loadCliConfig(
+        settings,
+        'test-session',
+        argv as unknown as CliArgs,
+        {
+          cwd: process.cwd(),
+        },
+      );
+
+      expect(config.getModel()).toBe('true');
+    });
+  });
+
   describe('Proxy configuration', () => {
     const originalProxyEnv: { [key: string]: string | undefined } = {};
     const proxyEnvVars = [
@@ -872,16 +990,14 @@ describe('loadCliConfig', () => {
   });
 
   it('should skip inaccessible workspace folders from GEMINI_CLI_IDE_WORKSPACE_PATH', async () => {
-    const resolveToRealPathSpy = vi
-      .spyOn(ServerConfig, 'resolveToRealPath')
-      .mockImplementation((p) => {
-        if (p.toString().includes('restricted')) {
-          const err = new Error('EACCES: permission denied');
-          (err as NodeJS.ErrnoException).code = 'EACCES';
-          throw err;
-        }
-        return p.toString();
-      });
+    vi.spyOn(ServerConfig, 'resolveToRealPath').mockImplementation((p) => {
+      if (p.toString().includes('restricted')) {
+        const err = new Error('EACCES: permission denied');
+        (err as NodeJS.ErrnoException).code = 'EACCES';
+        throw err;
+      }
+      return p.toString();
+    });
     vi.stubEnv(
       'GEMINI_CLI_IDE_WORKSPACE_PATH',
       ['/project/folderA', '/nonexistent/restricted/folder'].join(
@@ -895,8 +1011,6 @@ describe('loadCliConfig', () => {
     const dirs = config.getPendingIncludeDirectories();
     expect(dirs).toContain('/project/folderA');
     expect(dirs).not.toContain('/nonexistent/restricted/folder');
-
-    resolveToRealPathSpy.mockRestore();
   });
 
   it('should use default fileFilter options when unconfigured', async () => {
@@ -2217,51 +2331,6 @@ describe('loadCliConfig context management', () => {
       },
     });
     const config = await loadCliConfig(settings, 'test-session', argv);
-    expect(config.getContextManagementConfig()).toStrictEqual(
-      generalistProfile,
-    );
-    expect(config.isContextManagementEnabled()).toBe(true);
-  });
-
-  it('should be true when contextManagement is set to true in settings', async () => {
-    process.argv = ['node', 'script.js'];
-    const argv = await parseArguments(createTestMergedSettings());
-    const contextManagementConfig: Partial<ContextManagementConfig> = {
-      historyWindow: {
-        maxTokens: 100_000,
-        retainedTokens: 50_000,
-      },
-      messageLimits: {
-        normalMaxTokens: 1000,
-        retainedMaxTokens: 10_000,
-        normalizationHeadRatio: 0.25,
-      },
-      tools: {
-        distillation: {
-          maxOutputTokens: 10_000,
-          summarizationThresholdTokens: 15_000,
-        },
-        outputMasking: {
-          protectionThresholdTokens: 30_000,
-          minPrunableThresholdTokens: 10_000,
-          protectLatestTurn: false,
-        },
-      },
-    };
-    const settings = createTestMergedSettings({
-      experimental: {
-        contextManagement: true,
-      },
-      // The type of numbers is being inferred strangely, and so we have to cast
-      // to `any` here.
-      // eslint-disable-next-line @typescript-eslint/no-explicit-any
-      contextManagement: contextManagementConfig as any,
-    });
-    const config = await loadCliConfig(settings, 'test-session', argv);
-    expect(config.getContextManagementConfig()).toStrictEqual({
-      enabled: true,
-      ...contextManagementConfig,
-    });
     expect(config.isContextManagementEnabled()).toBe(true);
   });
 });
@@ -3225,7 +3294,7 @@ describe('Output format', () => {
   it('should error on invalid --output-format argument', async () => {
     process.argv = ['node', 'script.js', '--output-format', 'invalid'];
 
-    const mockExit = vi.spyOn(process, 'exit').mockImplementation(() => {
+    vi.spyOn(process, 'exit').mockImplementation(() => {
       throw new Error('process.exit called');
     });
 
@@ -3243,10 +3312,6 @@ describe('Output format', () => {
       expect.stringContaining('Invalid values:'),
     );
     expect(mockConsoleError).toHaveBeenCalled();
-
-    mockExit.mockRestore();
-    mockConsoleError.mockRestore();
-    debugErrorSpy.mockRestore();
   });
 });
 
@@ -3277,13 +3342,11 @@ describe('parseArguments with positional prompt', () => {
       'test prompt',
     ];
 
-    const mockExit = vi.spyOn(process, 'exit').mockImplementation(() => {
+    vi.spyOn(process, 'exit').mockImplementation(() => {
       throw new Error('process.exit called');
     });
 
-    const mockConsoleError = vi
-      .spyOn(console, 'error')
-      .mockImplementation(() => {});
+    vi.spyOn(console, 'error').mockImplementation(() => {});
     const debugErrorSpy = vi
       .spyOn(debugLogger, 'error')
       .mockImplementation(() => {});
@@ -3297,10 +3360,6 @@ describe('parseArguments with positional prompt', () => {
         'Cannot use both a positional prompt and the --prompt (-p) flag together',
       ),
     );
-
-    mockExit.mockRestore();
-    mockConsoleError.mockRestore();
-    debugErrorSpy.mockRestore();
   });
 
   it('should correctly parse a positional prompt to query field', async () => {

packages/cli/src/config/config.ts

@@ -48,7 +48,6 @@ import {
   type HookEventName,
   type OutputFormat,
   detectIdeFromEnv,
-  generalistProfile,
 } from '@google/gemini-cli-core';
 import {
   type Settings,
@@ -97,6 +96,7 @@ export interface CliArgs {
   extensions: string[] | undefined;
   listExtensions: boolean | undefined;
   resume: string | typeof RESUME_LATEST | undefined;
+  sessionId: string | undefined;
   listSessions: boolean | undefined;
   deleteSession: string | undefined;
   includeDirectories: string[] | undefined;
@@ -238,6 +238,10 @@ export async function parseArguments(
         ? query.length > 0
         : !!query;
 
+      if (argv['resume'] !== undefined && argv['session-id'] !== undefined) {
+        return 'Cannot use both --resume (-r) and --session-id together';
+      }
+
       if (argv['prompt'] && hasPositionalQuery) {
         return 'Cannot use both a positional prompt and the --prompt (-p) flag together';
       }
@@ -407,6 +411,25 @@ export async function parseArguments(
             return trimmed;
           },
         })
+        .option('session-id', {
+          type: 'string',
+          nargs: 1,
+          description: 'Start a new session with a manually provided UUID.',
+          coerce: (value: string): string => {
+            const trimmed = value.trim();
+            if (!trimmed) {
+              throw new Error('The --session-id option cannot be empty.');
+            }
+            if (!/^[a-zA-Z0-9-_]+$/.test(trimmed)) {
+              throw new Error(
+                'Invalid session ID "' +
+                  trimmed +
+                  '": Only alphanumeric characters, dashes, and underscores are allowed.',
+              );
+            }
+            return trimmed;
+          },
+        })
         .option('list-sessions', {
           type: 'boolean',
           description:
@@ -818,9 +841,16 @@ export async function loadCliConfig(
   );
 
   const defaultModel = PREVIEW_GEMINI_MODEL_AUTO;
-  const specifiedModel =
+  const rawModel =
     argv.model || process.env['GEMINI_MODEL'] || settings.model?.name;
 
+  // Ensure specifiedModel is a string (e.g. if yargs parsed multiple --model as an array)
+  const specifiedModel = Array.isArray(rawModel)
+    ? String(rawModel.at(-1) ?? '').trim() || ''
+    : rawModel === undefined
+      ? undefined
+      : String(rawModel ?? '').trim() || '';
+
   const resolvedModel =
     specifiedModel === GEMINI_MODEL_ALIAS_AUTO
       ? defaultModel
@@ -904,14 +934,19 @@ export async function loadCliConfig(
     }
   }
 
-  const useGeneralistProfile =
-    settings.experimental?.generalistProfile ?? false;
-  const useContextManagement =
-    settings.experimental?.contextManagement ?? false;
+  // TODO(joshualitt): Clean this up alongside removal of the legacy config.
+  let profileSelector: string | undefined = undefined;
+  if (settings.experimental?.stressTestProfile) {
+    profileSelector = 'stressTestProfile';
+  } else if (
+    settings.experimental?.generalistProfile ||
+    settings.experimental?.contextManagement
+  ) {
+    profileSelector = 'generalistProfile';
+  }
+
   const contextManagement = {
-    ...(useGeneralistProfile ? generalistProfile : {}),
-    ...(useContextManagement ? settings?.contextManagement : {}),
-    enabled: useContextManagement || useGeneralistProfile,
+    enabled: !!profileSelector,
   };
 
   return new Config({
@@ -935,6 +970,7 @@ export async function loadCliConfig(
     worktreeSettings,
 
     coreTools: settings.tools?.core || undefined,
+    experimentalContextManagementConfig: profileSelector,
     allowedTools: allowedTools.length > 0 ? allowedTools : undefined,
     policyEngineConfig,
     policyUpdateConfirmationRequest,
@@ -1000,6 +1036,7 @@ export async function loadCliConfig(
     enableExtensionReloading: settings.experimental?.extensionReloading,
     enableAgents: settings.experimental?.enableAgents,
     plan: settings.general?.plan?.enabled ?? true,
+    voiceMode: settings.experimental?.voiceMode,
     tracker: settings.experimental?.taskTracker,
     directWebFetch: settings.experimental?.directWebFetch,
     planSettings: settings.general?.plan?.directory

packages/cli/src/config/extension-manager-agents.test.ts

@@ -42,10 +42,12 @@ describe('ExtensionManager agents loading', () => {
 
   beforeEach(() => {
     vi.clearAllMocks();
+    vi.stubEnv('ANTIGRAVITY_CLI_ALIAS', '');
     vi.spyOn(debugLogger, 'warn').mockImplementation(() => {});
 
     tempDir = fs.mkdtempSync(path.join(os.tmpdir(), 'gemini-test-agents-'));
     mockHomedir.mockReturnValue(tempDir);
+    vi.stubEnv('GEMINI_CLI_HOME', tempDir);
 
     // Create the extensions directory that ExtensionManager expects
     extensionsDir = path.join(tempDir, '.gemini', EXTENSIONS_DIRECTORY_NAME);

packages/cli/src/config/extension-manager-hydration.test.ts

@@ -48,11 +48,13 @@ describe('ExtensionManager hydration', () => {
 
   beforeEach(() => {
     vi.clearAllMocks();
+    vi.stubEnv('ANTIGRAVITY_CLI_ALIAS', '');
     vi.spyOn(coreEvents, 'emitFeedback');
     vi.spyOn(debugLogger, 'debug').mockImplementation(() => {});
 
     tempDir = fs.mkdtempSync(path.join(os.tmpdir(), 'gemini-test-'));
     mockHomedir.mockReturnValue(tempDir);
+    vi.stubEnv('GEMINI_CLI_HOME', tempDir);
 
     // Create the extensions directory that ExtensionManager expects
     extensionsDir = path.join(tempDir, '.gemini', EXTENSIONS_DIRECTORY_NAME);

packages/cli/src/config/settings-validation.test.ts

@@ -13,6 +13,7 @@ import {
   settingsZodSchema,
 } from './settings-validation.js';
 import { z } from 'zod';
+import { type Settings } from './settingsSchema.js';
 
 describe('settings-validation', () => {
   describe('validateSettings', () => {
@@ -298,6 +299,117 @@ describe('settings-validation', () => {
         expect(issue).toBeDefined();
       }
     });
+
+    it('should accept customThemes with text.response color override', () => {
+      // Regression test for #25610: `response` is a documented and
+      // implemented color override for model responses (see
+      // packages/cli/src/ui/themes/theme.ts and semantic-tokens.ts),
+      // but was missing from the CustomTheme validation schema.
+      const validSettings = {
+        ui: {
+          theme: 'LimeWhite',
+          customThemes: {
+            LimeWhite: {
+              type: 'custom',
+              name: 'LimeWhite',
+              text: {
+                primary: '#00FF00',
+                response: '#FFFFFF',
+                secondary: '#a0a0a0',
+                accent: '#00FF00',
+              },
+            },
+          },
+        },
+      };
+
+      const result = validateSettings(validSettings);
+      expect(result.success).toBe(true);
+    });
+
+    describe('type casting', () => {
+      it('should cast "true" and "false" strings to booleans', () => {
+        const settings = {
+          ui: {
+            autoThemeSwitching: 'true',
+            hideWindowTitle: 'false',
+          },
+        };
+
+        const result = validateSettings(settings);
+        expect(result.success).toBe(true);
+        const data = result.data as Settings;
+        expect(data.ui?.autoThemeSwitching).toBe(true);
+        expect(data.ui?.hideWindowTitle).toBe(false);
+      });
+
+      it('should cast boolean strings case-insensitively', () => {
+        const settings = {
+          ui: {
+            autoThemeSwitching: 'TRUE',
+            hideWindowTitle: 'fAlSe',
+          },
+        };
+
+        const result = validateSettings(settings);
+        expect(result.success).toBe(true);
+        const data = result.data as Settings;
+        expect(data.ui?.autoThemeSwitching).toBe(true);
+        expect(data.ui?.hideWindowTitle).toBe(false);
+      });
+
+      it('should cast numeric strings to numbers', () => {
+        const settings = {
+          model: {
+            maxSessionTurns: '42',
+            compressionThreshold: '0.5',
+          },
+        };
+
+        const result = validateSettings(settings);
+        expect(result.success).toBe(true);
+        const data = result.data as Settings;
+        expect(data.model?.maxSessionTurns).toBe(42);
+        expect(data.model?.compressionThreshold).toBe(0.5);
+      });
+
+      it('should reject invalid castable strings', () => {
+        const settings = {
+          ui: {
+            autoThemeSwitching: 'not-a-boolean',
+          },
+          model: {
+            maxSessionTurns: 'not-a-number',
+          },
+        };
+
+        const result = validateSettings(settings);
+        expect(result.success).toBe(false);
+        expect(result.error?.issues).toHaveLength(2);
+        expect(result.error?.issues[0].message).toContain(
+          'Expected boolean, received string',
+        );
+        expect(result.error?.issues[1].message).toContain(
+          'Expected number, received string',
+        );
+      });
+
+      it('should cast strings to booleans/numbers in shared definitions (refs)', () => {
+        const settings = {
+          mcpServers: {
+            'test-server': {
+              command: 'node',
+              trust: 'true', // from boolean ref
+            },
+          },
+        };
+
+        const result = validateSettings(settings);
+        expect(result.success).toBe(true);
+        const data = result.data as Settings;
+        expect(data.mcpServers?.['test-server'].trust).toBe(true);
+      });
+    });
   });
 
   describe('formatValidationError', () => {

packages/cli/src/config/settings-validation.ts

@@ -25,10 +25,10 @@ function buildZodSchemaFromJsonSchema(def: any): z.ZodTypeAny {
   if (def.type === 'string') {
     // eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
     if (def.enum) return z.enum(def.enum as [string, ...string[]]);
-    return z.string();
+    return buildPrimitiveSchema('string');
   }
-  if (def.type === 'number') return z.number();
-  if (def.type === 'boolean') return z.boolean();
+  if (def.type === 'number') return buildPrimitiveSchema('number');
+  if (def.type === 'boolean') return buildPrimitiveSchema('boolean');
 
   if (def.type === 'array') {
     if (def.items) {
@@ -133,9 +133,22 @@ function buildPrimitiveSchema(
     case 'string':
       return z.string();
     case 'number':
-      return z.number();
+      return z.preprocess((val) => {
+        if (typeof val === 'string' && val.trim() !== '') {
+          const num = Number(val);
+          if (!isNaN(num)) return num;
+        }
+        return val;
+      }, z.number());
     case 'boolean':
-      return z.boolean();
+      return z.preprocess((val) => {
+        if (typeof val === 'string') {
+          const lower = val.toLowerCase();
+          if (lower === 'true') return true;
+          if (lower === 'false') return false;
+        }
+        return val;
+      }, z.boolean());
     default:
       return z.unknown();
   }
@@ -160,7 +173,9 @@ function buildZodSchemaFromDefinition(
   if (definition.ref === 'TelemetrySettings') {
     const objectSchema = REF_SCHEMAS['TelemetrySettings'];
     if (objectSchema) {
-      return z.union([z.boolean(), objectSchema]).optional();
+      return z
+        .union([buildPrimitiveSchema('boolean'), objectSchema])
+        .optional();
     }
   }
 

packages/cli/src/config/settings.test.ts

@@ -479,6 +479,137 @@ describe('Settings Loading and Merging', () => {
       expect(settings.merged.security?.folderTrust?.enabled).toBe(false); // Workspace setting should be used
     });
 
+    it('should resolve environment variables and cast them to correct types before validation', () => {
+      vi.stubEnv('TEST_AUTO_THEME', 'false');
+      vi.stubEnv('TEST_MAX_TURNS', '15');
+
+      (mockFsExistsSync as Mock).mockImplementation(
+        (p: fs.PathLike) =>
+          path.normalize(p.toString()) === path.normalize(USER_SETTINGS_PATH),
+      );
+      (fs.readFileSync as Mock).mockImplementation(
+        (p: fs.PathOrFileDescriptor) => {
+          if (
+            path.normalize(p.toString()) === path.normalize(USER_SETTINGS_PATH)
+          ) {
+            return JSON.stringify({
+              ui: { autoThemeSwitching: '$TEST_AUTO_THEME' },
+              model: { maxSessionTurns: '$TEST_MAX_TURNS' },
+            });
+          }
+          return '{}';
+        },
+      );
+
+      const settings = loadSettings(MOCK_WORKSPACE_DIR);
+
+      expect(settings.merged.ui.autoThemeSwitching).toBe(false);
+      expect(settings.merged.model.maxSessionTurns).toBe(15);
+      expect(settings.errors).toHaveLength(0);
+    });
+
+    it('should use default values from environment variable placeholders', () => {
+      vi.stubEnv('TEST_AUTO_THEME', ''); // Should trigger default
+      delete process.env['TEST_AUTO_THEME'];
+
+      (mockFsExistsSync as Mock).mockImplementation(
+        (p: fs.PathLike) =>
+          path.normalize(p.toString()) === path.normalize(USER_SETTINGS_PATH),
+      );
+      (fs.readFileSync as Mock).mockImplementation(
+        (p: fs.PathOrFileDescriptor) => {
+          if (
+            path.normalize(p.toString()) === path.normalize(USER_SETTINGS_PATH)
+          ) {
+            return JSON.stringify({
+              ui: { autoThemeSwitching: '${TEST_AUTO_THEME:-true}' },
+            });
+          }
+          return '{}';
+        },
+      );
+
+      const settings = loadSettings(MOCK_WORKSPACE_DIR);
+
+      expect(settings.merged.ui.autoThemeSwitching).toBe(true);
+      expect(settings.errors).toHaveLength(0);
+    });
+
+    it('should record validation errors if expansion result is invalid', () => {
+      vi.stubEnv('TEST_MAX_TURNS', 'not-a-number');
+
+      (mockFsExistsSync as Mock).mockImplementation(
+        (p: fs.PathLike) =>
+          path.normalize(p.toString()) === path.normalize(USER_SETTINGS_PATH),
+      );
+      (fs.readFileSync as Mock).mockImplementation(
+        (p: fs.PathOrFileDescriptor) => {
+          if (
+            path.normalize(p.toString()) === path.normalize(USER_SETTINGS_PATH)
+          ) {
+            return JSON.stringify({
+              model: { maxSessionTurns: '$TEST_MAX_TURNS' },
+            });
+          }
+          return '{}';
+        },
+      );
+
+      const settings = loadSettings(MOCK_WORKSPACE_DIR);
+
+      expect(settings.errors.length).toBeGreaterThan(0);
+      expect(settings.errors[0].message).toContain(
+        'Expected number, received string',
+      );
+      // Should fall back to the expanded string value
+      expect(settings.merged.model.maxSessionTurns).toBe('not-a-number');
+    });
+
+    it('should preserve environment variable placeholders on save', () => {
+      vi.stubEnv('TEST_AUTO_THEME', 'true');
+      const placeholder = '${TEST_AUTO_THEME:-false}';
+
+      (mockFsExistsSync as Mock).mockImplementation(
+        (p: fs.PathLike) =>
+          path.normalize(p.toString()) === path.normalize(USER_SETTINGS_PATH),
+      );
+      (fs.readFileSync as Mock).mockImplementation(
+        (p: fs.PathOrFileDescriptor) => {
+          if (
+            path.normalize(p.toString()) === path.normalize(USER_SETTINGS_PATH)
+          ) {
+            return JSON.stringify({
+              ui: { autoThemeSwitching: placeholder },
+            });
+          }
+          return '{}';
+        },
+      );
+
+      // Load settings - this will expand the placeholder for runtime use
+      const loaded = loadSettings(MOCK_WORKSPACE_DIR);
+      expect(loaded.merged.ui.autoThemeSwitching).toBe(true);
+
+      // Verify that the original settings for the user scope still have the placeholder
+      const userFile = loaded.forScope(SettingScope.User);
+      expect(userFile.originalSettings.ui?.autoThemeSwitching).toBe(
+        placeholder,
+      );
+
+      // Save settings - this should use the originalSettings (with placeholders)
+      const mockUpdate = vi.mocked(updateSettingsFilePreservingFormat);
+      saveSettings(userFile);
+
+      expect(mockUpdate).toHaveBeenCalledWith(
+        USER_SETTINGS_PATH,
+        expect.objectContaining({
+          ui: expect.objectContaining({
+            autoThemeSwitching: placeholder,
+          }),
+        }),
+      );
+    });
+
     it('should use system folderTrust over user setting', () => {
       (mockFsExistsSync as Mock).mockReturnValue(true);
       const userSettingsContent = {

packages/cli/src/config/settings.ts

@@ -673,7 +673,9 @@ function _doLoadSettings(workspaceDir: string): LoadedSettings {
   const storage = new Storage(workspaceDir);
   const workspaceSettingsPath = storage.getWorkspaceSettingsPath();
 
-  const load = (filePath: string): { settings: Settings; rawJson?: string } => {
+  const load = (
+    filePath: string,
+  ): { settings: Settings; rawSettings: Settings; rawJson?: string } => {
     try {
       if (fs.existsSync(filePath)) {
         const content = fs.readFileSync(filePath, 'utf-8');
@@ -689,14 +691,19 @@ function _doLoadSettings(workspaceDir: string): LoadedSettings {
             path: filePath,
             severity: 'error',
           });
-          return { settings: {} };
+          return { settings: {}, rawSettings: {} };
         }
 
         // eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
         const settingsObject = rawSettings as Record<string, unknown>;
 
-        // Validate settings structure with Zod
-        const validationResult = validateSettings(settingsObject);
+        // Expand environment variables
+        const expandedSettings = resolveEnvVarsInObject(
+          settingsObject as Settings,
+        );
+
+        // Validate settings structure with Zod after environment variable expansion
+        const validationResult = validateSettings(expandedSettings);
         if (!validationResult.success && validationResult.error) {
           const errorMessage = formatValidationError(
             validationResult.error,
@@ -707,9 +714,22 @@ function _doLoadSettings(workspaceDir: string): LoadedSettings {
             path: filePath,
             severity: 'warning',
           });
+          return {
+            settings: expandedSettings,
+            rawSettings: settingsObject as Settings,
+            rawJson: content,
+          };
         }
 
-        return { settings: settingsObject as Settings, rawJson: content };
+        // Return the successfully cast and validated data
+        return {
+          // Since we've successfully validated expandedSettings against settingsZodSchema,
+          // it's safe to cast the resulting data to the Settings type.
+          // eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
+          settings: (validationResult.data as Settings) ?? expandedSettings,
+          rawSettings: settingsObject as Settings,
+          rawJson: content,
+        };
       }
     } catch (error: unknown) {
       settingsErrors.push({
@@ -718,33 +738,40 @@ function _doLoadSettings(workspaceDir: string): LoadedSettings {
         severity: 'error',
       });
     }
-    return { settings: {} };
+    return { settings: {}, rawSettings: {} };
   };
 
   const systemResult = load(systemSettingsPath);
   const systemDefaultsResult = load(systemDefaultsPath);
   const userResult = load(USER_SETTINGS_PATH);
 
-  let workspaceResult: { settings: Settings; rawJson?: string } = {
+  let workspaceResult: {
+    settings: Settings;
+    rawSettings: Settings;
+    rawJson?: string;
+  } = {
     settings: {} as Settings,
+    rawSettings: {} as Settings,
     rawJson: undefined,
   };
   if (!storage.isWorkspaceHomeDir()) {
     workspaceResult = load(workspaceSettingsPath);
   }
 
-  const systemOriginalSettings = structuredClone(systemResult.settings);
+  const systemOriginalSettings = structuredClone(systemResult.rawSettings);
   const systemDefaultsOriginalSettings = structuredClone(
-    systemDefaultsResult.settings,
+    systemDefaultsResult.rawSettings,
+  );
+  const userOriginalSettings = structuredClone(userResult.rawSettings);
+  const workspaceOriginalSettings = structuredClone(
+    workspaceResult.rawSettings,
   );
-  const userOriginalSettings = structuredClone(userResult.settings);
-  const workspaceOriginalSettings = structuredClone(workspaceResult.settings);
 
-  // Environment variables for runtime use
-  systemSettings = resolveEnvVarsInObject(systemResult.settings);
-  systemDefaultSettings = resolveEnvVarsInObject(systemDefaultsResult.settings);
-  userSettings = resolveEnvVarsInObject(userResult.settings);
-  workspaceSettings = resolveEnvVarsInObject(workspaceResult.settings);
+  // Environment variables for runtime use are already resolved and validated in load()
+  systemSettings = systemResult.settings;
+  systemDefaultSettings = systemDefaultsResult.settings;
+  userSettings = userResult.settings;
+  workspaceSettings = workspaceResult.settings;
 
   // Support legacy theme names
   if (userSettings.ui?.theme === 'VS') {

packages/cli/src/config/settingsSchema.ts

@@ -2061,6 +2061,87 @@ const SETTINGS_SCHEMA = {
         description: 'Enable access to Gemma 4 models (experimental).',
         showInDialog: true,
       },
+      voiceMode: {
+        type: 'boolean',
+        label: 'Voice Mode',
+        category: 'Experimental',
+        requiresRestart: false,
+        default: false,
+        description:
+          'Enable experimental voice dictation and commands (/voice, /voice model).',
+        showInDialog: true,
+      },
+      voice: {
+        type: 'object',
+        label: 'Voice',
+        category: 'Experimental',
+        requiresRestart: false,
+        default: {},
+        description: 'Settings for voice mode and transcription.',
+        showInDialog: false,
+        properties: {
+          activationMode: {
+            type: 'enum',
+            label: 'Voice Activation Mode',
+            category: 'Experimental',
+            requiresRestart: false,
+            default: 'push-to-talk',
+            description: 'How to trigger voice recording with the Space key.',
+            showInDialog: true,
+            options: [
+              { value: 'push-to-talk', label: 'Push-To-Talk (Hold Space)' },
+              { value: 'toggle', label: 'Toggle (Press Space to start/stop)' },
+            ],
+          },
+          backend: {
+            type: 'enum',
+            label: 'Voice Transcription Backend',
+            category: 'Experimental',
+            requiresRestart: false,
+            default: 'gemini-live',
+            description: 'The backend to use for voice transcription.',
+            showInDialog: true,
+            options: [
+              { value: 'gemini-live', label: 'Gemini Live API (Cloud)' },
+              { value: 'whisper', label: 'Whisper (Local)' },
+            ],
+          },
+          whisperModel: {
+            type: 'enum',
+            label: 'Whisper Model',
+            category: 'Experimental',
+            requiresRestart: false,
+            default: 'ggml-base.en.bin',
+            description: 'The Whisper model to use for local transcription.',
+            showInDialog: true,
+            options: [
+              { value: 'ggml-tiny.en.bin', label: 'Tiny (EN) - Fast (~75MB)' },
+              {
+                value: 'ggml-base.en.bin',
+                label: 'Base (EN) - Balanced (~142MB)',
+              },
+              {
+                value: 'ggml-large-v3-turbo-q5_0.bin',
+                label: 'Large v3 Turbo (Q5_0) - High Accuracy (~547MB)',
+              },
+              {
+                value: 'ggml-large-v3-turbo-q8_0.bin',
+                label: 'Large v3 Turbo (Q8_0) - Max Accuracy (~834MB)',
+              },
+            ],
+          },
+          stopGracePeriodMs: {
+            type: 'number',
+            label: 'Voice Stop Grace Period (ms)',
+            category: 'Experimental',
+            requiresRestart: false,
+            default: 1000,
+            description:
+              'How long to wait for final transcription after stopping recording.',
+            showInDialog: true,
+          },
+        },
+      },
       adk: {
         type: 'object',
         label: 'ADK',
@@ -2307,6 +2388,17 @@ const SETTINGS_SCHEMA = {
           'Disable the built-in save_memory tool and let the main agent persist project context by editing markdown files directly with edit/write_file. Route facts across four tiers: team-shared conventions go to project GEMINI.md files, project-specific personal notes go to the per-project private memory folder (MEMORY.md as index + sibling .md files for detail), and cross-project personal preferences go to the global ~/.gemini/GEMINI.md (the only file under ~/.gemini/ that the agent can edit — settings, credentials, etc. remain off-limits). Set to false to fall back to the legacy save_memory tool.',
         showInDialog: true,
       },
+      stressTestProfile: {
+        type: 'boolean',
+        label:
+          'Use the stress test profile to aggressively trigger context management.',
+        category: 'Experimental',
+        requiresRestart: true,
+        default: false,
+        description:
+          'Significantly lowers token limits to force early garbage collection and distillation for testing purposes.',
+        showInDialog: false,
+      },
       autoMemory: {
         type: 'boolean',
         label: 'Auto Memory',
@@ -3197,6 +3289,7 @@ export const SETTINGS_SCHEMA_DEFINITIONS: Record<
           secondary: { type: 'string' },
           link: { type: 'string' },
           accent: { type: 'string' },
+          response: { type: 'string' },
         },
       },
       background: {

packages/cli/src/gemini.test.tsx

@@ -20,6 +20,7 @@ import {
   validateDnsResolutionOrder,
   startInteractiveUI,
   getNodeMemoryArgs,
+  resolveSessionId,
 } from './gemini.js';
 import {
   loadCliConfig,
@@ -47,10 +48,13 @@ import {
   debugLogger,
   coreEvents,
   AuthType,
+  ExitCodes,
 } from '@google/gemini-cli-core';
 import { act } from 'react';
 import { type InitializationResult } from './core/initializer.js';
 import { runNonInteractive } from './nonInteractiveCli.js';
+import { SessionSelector, SessionError } from './utils/sessionUtils.js';
+
 // Hoisted constants and mocks
 const performance = vi.hoisted(() => ({
   now: vi.fn(),
@@ -548,6 +552,7 @@ describe('gemini.tsx main function kitty protocol', () => {
       screenReader: undefined,
       useWriteTodos: undefined,
       resume: undefined,
+      sessionId: undefined,
       listSessions: undefined,
       deleteSession: undefined,
       outputFormat: undefined,
@@ -607,6 +612,7 @@ describe('gemini.tsx main function kitty protocol', () => {
       screenReader: undefined,
       useWriteTodos: undefined,
       resume: undefined,
+      sessionId: undefined,
       listSessions: undefined,
       deleteSession: undefined,
       outputFormat: undefined,
@@ -822,7 +828,6 @@ describe('gemini.tsx main function kitty protocol', () => {
   });
 
   it('should handle session selector error', async () => {
-    const { SessionSelector } = await import('./utils/sessionUtils.js');
     vi.mocked(SessionSelector).mockImplementation(
       () =>
         ({
@@ -879,9 +884,6 @@ describe('gemini.tsx main function kitty protocol', () => {
   });
 
   it('should start normally with a warning when no sessions found for resume', async () => {
-    const { SessionSelector, SessionError } = await import(
-      './utils/sessionUtils.js'
-    );
     vi.mocked(SessionSelector).mockImplementation(
       () =>
         ({
@@ -1056,6 +1058,63 @@ describe('gemini.tsx main function kitty protocol', () => {
   });
 });
 
+describe('resolveSessionId', () => {
+  it('should return a new session ID when neither resume nor sessionId is provided', async () => {
+    const { sessionId, resumedSessionData } = await resolveSessionId(
+      undefined,
+      undefined,
+    );
+    expect(sessionId).toBeDefined();
+    expect(resumedSessionData).toBeUndefined();
+  });
+
+  it('should exit with FATAL_INPUT_ERROR when sessionId already exists', async () => {
+    vi.mocked(SessionSelector).mockImplementation(
+      () =>
+        ({
+          sessionExists: vi.fn().mockResolvedValue(true),
+        }) as unknown as InstanceType<typeof SessionSelector>,
+    );
+
+    const emitFeedbackSpy = vi.spyOn(coreEvents, 'emitFeedback');
+    const processExitSpy = vi
+      .spyOn(process, 'exit')
+      .mockImplementation((code) => {
+        throw new MockProcessExitError(code);
+      });
+
+    try {
+      await resolveSessionId(undefined, 'existing-id');
+    } catch (e) {
+      if (!(e instanceof MockProcessExitError)) throw e;
+    }
+
+    expect(emitFeedbackSpy).toHaveBeenCalledWith(
+      'error',
+      expect.stringContaining('Session ID "existing-id" already exists'),
+    );
+    expect(processExitSpy).toHaveBeenCalledWith(ExitCodes.FATAL_INPUT_ERROR);
+
+    emitFeedbackSpy.mockRestore();
+    processExitSpy.mockRestore();
+  });
+
+  it('should return provided sessionId when it does not exist', async () => {
+    vi.mocked(SessionSelector).mockImplementation(
+      () =>
+        ({
+          sessionExists: vi.fn().mockResolvedValue(false),
+        }) as unknown as InstanceType<typeof SessionSelector>,
+    );
+    const { sessionId, resumedSessionData } = await resolveSessionId(
+      undefined,
+      'new-id',
+    );
+    expect(sessionId).toBe('new-id');
+    expect(resumedSessionData).toBeUndefined();
+  });
+});
+
 describe('gemini.tsx main function exit codes', () => {
   let originalEnvNoRelaunch: string | undefined;
   let originalIsTTY: boolean | undefined;

packages/cli/src/gemini.tsx

@@ -76,7 +76,7 @@ import {
   type InitializationResult,
 } from './core/initializer.js';
 import { validateAuthMethod } from './config/auth.js';
-import { runAcpClient } from './acp/acpClient.js';
+import { runAcpClient } from './acp/acpStdioTransport.js';
 import { validateNonInteractiveAuth } from './validateNonInterActiveAuth.js';
 import { appEvents, AppEvent } from './utils/events.js';
 import { SessionError, SessionSelector } from './utils/sessionUtils.js';
@@ -85,7 +85,6 @@ import { relaunchOnExitCode } from './utils/relaunch.js';
 import { loadSandboxConfig } from './config/sandboxConfig.js';
 import { deleteSession, listSessions } from './utils/sessions.js';
 import { createPolicyUpdater } from './config/policy.js';
-import { isAlternateBufferEnabled } from './ui/hooks/useAlternateBuffer.js';
 
 import { setupTerminalAndTheme } from './utils/terminalTheme.js';
 import { runDeferredCommand } from './deferred.js';
@@ -191,21 +190,38 @@ ${reason.stack}`
   });
 }
 
-export async function resolveSessionId(resumeArg: string | undefined): Promise<{
+export async function resolveSessionId(
+  resumeArg: string | undefined,
+  sessionIdArg?: string | undefined,
+): Promise<{
   sessionId: string;
   resumedSessionData?: ResumedSessionData;
 }> {
-  if (!resumeArg) {
+  if (!resumeArg && !sessionIdArg) {
     return { sessionId: createSessionId() };
   }
 
   const storage = new Storage(process.cwd());
   await storage.initialize();
 
+  const sessionSelector = new SessionSelector(storage);
+
+  if (sessionIdArg) {
+    if (await sessionSelector.sessionExists(sessionIdArg)) {
+      coreEvents.emitFeedback(
+        'error',
+        `Error starting session: Session ID "${sessionIdArg}" already exists. Use --resume to resume it, or provide a different ID.`,
+      );
+      await runExitCleanup();
+      process.exit(ExitCodes.FATAL_INPUT_ERROR);
+    }
+    return { sessionId: sessionIdArg };
+  }
+
   try {
-    const { sessionData, sessionPath } = await new SessionSelector(
-      storage,
-    ).resolveSession(resumeArg);
+    const { sessionData, sessionPath } = await sessionSelector.resolveSession(
+      resumeArg!,
+    );
     return {
       sessionId: sessionData.sessionId,
       resumedSessionData: { conversation: sessionData, filePath: sessionPath },
@@ -319,7 +335,10 @@ export async function main() {
 
   const argv = await argvPromise;
 
-  const { sessionId, resumedSessionData } = await resolveSessionId(argv.resume);
+  const { sessionId, resumedSessionData } = await resolveSessionId(
+    argv.resume,
+    argv.sessionId,
+  );
 
   if (
     (argv.allowedTools && argv.allowedTools.length > 0) ||
@@ -358,15 +377,14 @@ export async function main() {
 
   const isDebugMode = cliConfig.isDebugMode(argv);
   const consolePatcher = new ConsolePatcher({
-    stderr: true,
-    interactive: isHeadlessMode() ? false : true,
+    stderr: argv.isCommand ? false : true,
+    interactive: isHeadlessMode() && !argv.isCommand ? false : true,
     debugMode: isDebugMode,
     onNewMessage: (msg) => {
       coreEvents.emitConsoleLog(msg.type, msg.content);
     },
   });
   consolePatcher.patch();
-  registerCleanup(consolePatcher.cleanup);
 
   dns.setDefaultResultOrder(
     validateDnsResolutionOrder(settings.merged.advanced.dnsResolutionOrder),
@@ -392,6 +410,7 @@ export async function main() {
   const partialConfig = await loadCliConfig(settings.merged, sessionId, argv, {
     projectHooks: settings.workspace.settings.hooks,
   });
+
   adminControlsListner.setConfig(partialConfig);
 
   // Refresh auth to fetch remote admin settings from CCPA and before entering
@@ -549,6 +568,12 @@ export async function main() {
       await config.getHookSystem()?.fireSessionEndEvent(SessionEndReason.Exit);
     });
 
+    // Register ConsolePatcher cleanup last to ensure logs from shutdown hooks
+    // are correctly redirected to stderr (especially for non-interactive JSON output).
+    if (!config.getAcpMode()) {
+      registerCleanup(consolePatcher.cleanup);
+    }
+
     // Launch cleanup expired sessions as a background task
     cleanupExpiredSessions(config, settings.merged).catch((e) => {
       debugLogger.error('Failed to cleanup expired sessions:', e);
@@ -644,7 +669,7 @@ export async function main() {
 
     let input = config.getQuestion();
     const useAlternateBuffer = shouldEnterAlternateScreen(
-      isAlternateBufferEnabled(config),
+      config.getUseAlternateBuffer(),
       config.getScreenReader(),
     );
     const rawStartupWarnings = await rawStartupWarningsPromise;
@@ -786,20 +811,16 @@ export function initializeOutputListenersAndFlush() {
     if (coreEvents.listenerCount(CoreEvent.ConsoleLog) === 0) {
       coreEvents.on(CoreEvent.ConsoleLog, (payload: ConsoleLogPayload) => {
         if (payload.type === 'error' || payload.type === 'warn') {
-          writeToStderr(payload.content);
+          writeToStderr(payload.content + '\n');
         } else {
-          writeToStdout(payload.content);
+          writeToStderr(payload.content + '\n');
         }
       });
     }
 
     if (coreEvents.listenerCount(CoreEvent.UserFeedback) === 0) {
       coreEvents.on(CoreEvent.UserFeedback, (payload: UserFeedbackPayload) => {
-        if (payload.severity === 'error' || payload.severity === 'warning') {
-          writeToStderr(payload.message);
-        } else {
-          writeToStdout(payload.message);
-        }
+        writeToStderr(payload.message + '\n');
       });
     }
   }

packages/cli/src/gemini_cleanup.test.tsx

@@ -68,6 +68,13 @@ vi.mock('./config/settings.js', async (importOriginal) => {
   };
 });
 
+vi.mock('./ui/utils/ConsolePatcher.js', () => ({
+  ConsolePatcher: vi.fn().mockImplementation(() => ({
+    patch: vi.fn(),
+    cleanup: vi.fn(),
+  })),
+}));
+
 vi.mock('./config/config.js', () => ({
   loadCliConfig: vi.fn().mockResolvedValue({
     getSandbox: vi.fn(() => false),
@@ -150,6 +157,10 @@ vi.mock('./utils/cleanup.js', async (importOriginal) => {
   };
 });
 
+vi.mock('./acp/acpStdioTransport.js', () => ({
+  runAcpClient: vi.fn().mockResolvedValue(undefined),
+}));
+
 vi.mock('./zed-integration/zedIntegration.js', () => ({
   runZedIntegration: vi.fn().mockResolvedValue(undefined),
 }));
@@ -296,6 +307,120 @@ describe('gemini.tsx main function cleanup', () => {
     );
   });
 
+  it('should not register ConsolePatcher cleanup in ACP mode', async () => {
+    const { registerCleanup } = await import('./utils/cleanup.js');
+    const { ConsolePatcher } = await import('./ui/utils/ConsolePatcher.js');
+    const { loadCliConfig, parseArguments } = await import(
+      './config/config.js'
+    );
+    const { loadSettings } = await import('./config/settings.js');
+
+    vi.mocked(parseArguments).mockResolvedValue({
+      acp: true,
+      startupMessages: [],
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    } as any);
+
+    vi.mocked(loadSettings).mockReturnValue({
+      merged: {
+        tools: { allowed: [], exclude: [] },
+        advanced: { dnsResolutionOrder: 'ipv4first' },
+        security: { auth: { selectedType: 'google' } },
+        ui: { theme: 'default' },
+      },
+      workspace: { settings: {} },
+      errors: [],
+      subscribe: vi.fn(),
+      getSnapshot: vi.fn(),
+      setValue: vi.fn(),
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    } as any);
+
+    vi.mocked(loadCliConfig).mockResolvedValue(
+      buildMockConfig({
+        getAcpMode: () => true,
+      }),
+    );
+
+    let capturedCleanup: () => void;
+    vi.mocked(ConsolePatcher).mockImplementation(() => {
+      const instance = {
+        patch: vi.fn(),
+        cleanup: vi.fn(),
+      };
+      capturedCleanup = instance.cleanup;
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+      return instance as any;
+    });
+
+    await main();
+
+    const registeredFunctions = vi
+      .mocked(registerCleanup)
+      .mock.calls.map((call) => call[0]);
+    expect(registeredFunctions).not.toContain(capturedCleanup!);
+  });
+
+  it('should register ConsolePatcher cleanup in non-ACP mode', async () => {
+    const { registerCleanup } = await import('./utils/cleanup.js');
+    const { ConsolePatcher } = await import('./ui/utils/ConsolePatcher.js');
+    const { loadCliConfig, parseArguments } = await import(
+      './config/config.js'
+    );
+    const { loadSettings } = await import('./config/settings.js');
+
+    vi.mocked(parseArguments).mockResolvedValue({
+      acp: false,
+      query: 'test',
+      startupMessages: [],
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    } as any);
+
+    vi.mocked(loadSettings).mockReturnValue({
+      merged: {
+        tools: { allowed: [], exclude: [] },
+        advanced: { dnsResolutionOrder: 'ipv4first' },
+        security: { auth: { selectedType: 'google' } },
+        ui: { theme: 'default' },
+      },
+      workspace: { settings: {} },
+      errors: [],
+      subscribe: vi.fn(),
+      getSnapshot: vi.fn(),
+      setValue: vi.fn(),
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    } as any);
+
+    vi.mocked(loadCliConfig).mockResolvedValue(
+      buildMockConfig({
+        getAcpMode: () => false,
+        getQuestion: () => 'test',
+      }),
+    );
+
+    let capturedCleanup: () => void;
+    vi.mocked(ConsolePatcher).mockImplementation(() => {
+      const instance = {
+        patch: vi.fn(),
+        cleanup: vi.fn(),
+      };
+      capturedCleanup = instance.cleanup;
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+      return instance as any;
+    });
+
+    try {
+      await main();
+    } catch {
+      // Ignore errors from incomplete mocks in full main() execution
+    }
+
+    const registeredFunctions = vi
+      .mocked(registerCleanup)
+      .mock.calls.map((call) => call[0]);
+    expect(registeredFunctions).toContain(capturedCleanup!);
+  });
+
   function buildMockConfig(overrides: Partial<Config> = {}): Config {
     return {
       isInteractive: vi.fn(() => false),
@@ -319,7 +444,6 @@ describe('gemini.tsx main function cleanup', () => {
       getListExtensions: vi.fn(() => false),
       getListSessions: vi.fn(() => false),
       getDeleteSession: vi.fn(() => undefined),
-      getToolRegistry: vi.fn(),
       getExtensions: vi.fn(() => []),
       getModel: vi.fn(() => 'gemini-pro'),
       getEmbeddingModel: vi.fn(() => 'embedding-001'),

packages/cli/src/services/BuiltinCommandLoader.test.ts

@@ -170,6 +170,7 @@ describe('BuiltinCommandLoader', () => {
         getAllSkills: vi.fn().mockReturnValue([]),
         isAdminEnabled: vi.fn().mockReturnValue(true),
       }),
+      isVoiceModeEnabled: vi.fn().mockReturnValue(true),
       getContentGeneratorConfig: vi.fn().mockReturnValue({
         authType: 'other',
       }),
@@ -396,6 +397,7 @@ describe('BuiltinCommandLoader profile', () => {
         getAllSkills: vi.fn().mockReturnValue([]),
         isAdminEnabled: vi.fn().mockReturnValue(true),
       }),
+      isVoiceModeEnabled: vi.fn().mockReturnValue(true),
       getContentGeneratorConfig: vi.fn().mockReturnValue({
         authType: 'other',
       }),

packages/cli/src/services/BuiltinCommandLoader.ts

@@ -62,6 +62,7 @@ import { setupGithubCommand } from '../ui/commands/setupGithubCommand.js';
 import { terminalSetupCommand } from '../ui/commands/terminalSetupCommand.js';
 import { upgradeCommand } from '../ui/commands/upgradeCommand.js';
 import { gemmaStatusCommand } from '../ui/commands/gemmaStatusCommand.js';
+import { voiceCommand } from '../ui/commands/voiceCommand.js';
 
 /**
  * Loads the core, hard-coded slash commands that are an integral part
@@ -227,6 +228,7 @@ export class BuiltinCommandLoader implements ICommandLoader {
       vimCommand,
       setupGithubCommand,
       terminalSetupCommand,
+      ...(this.config?.isVoiceModeEnabled() ? [voiceCommand] : []),
       ...(this.config?.getContentGeneratorConfig()?.authType ===
       AuthType.LOGIN_WITH_GOOGLE
         ? [upgradeCommand]

packages/cli/src/test-utils/render.tsx

@@ -552,6 +552,8 @@ const mockUIActions: UIActions = {
   exitPrivacyNotice: vi.fn(),
   closeSettingsDialog: vi.fn(),
   closeModelDialog: vi.fn(),
+  openVoiceModelDialog: vi.fn(),
+  closeVoiceModelDialog: vi.fn(),
   openAgentConfigDialog: vi.fn(),
   closeAgentConfigDialog: vi.fn(),
   openPermissionsDialog: vi.fn(),
@@ -598,6 +600,7 @@ const mockUIActions: UIActions = {
   handleNewAgentsSelect: vi.fn(),
   getPreferredEditor: vi.fn(),
   clearAccountSuspension: vi.fn(),
+  setVoiceModeEnabled: vi.fn(),
 };
 
 import { type TextBuffer } from '../ui/components/shared/text-buffer.js';

packages/cli/src/ui/AppContainer.tsx

@@ -103,6 +103,7 @@ import { useQuotaAndFallback } from './hooks/useQuotaAndFallback.js';
 import { useEditorSettings } from './hooks/useEditorSettings.js';
 import { useSettingsCommand } from './hooks/useSettingsCommand.js';
 import { useModelCommand } from './hooks/useModelCommand.js';
+import { useVoiceModelCommand } from './hooks/useVoiceModelCommand.js';
 import { useSlashCommandProcessor } from './hooks/slashCommandProcessor.js';
 import { useVimMode } from './contexts/VimModeContext.js';
 import {
@@ -312,6 +313,7 @@ export const AppContainer = (props: AppContainerProps) => {
   );
 
   const [shellModeActive, setShellModeActive] = useState(false);
+  const [isVoiceModeEnabled, setVoiceModeEnabled] = useState(false);
   const [modelSwitchedFromQuotaError, setModelSwitchedFromQuotaError] =
     useState<boolean>(false);
   const [historyRemountKey, setHistoryRemountKey] = useState(0);
@@ -946,6 +948,12 @@ Logging in with Google... Restarting Gemini CLI to continue.
   const { isModelDialogOpen, openModelDialog, closeModelDialog } =
     useModelCommand();
 
+  const {
+    isVoiceModelDialogOpen,
+    openVoiceModelDialog,
+    closeVoiceModelDialog,
+  } = useVoiceModelCommand();
+
   const { toggleVimEnabled } = useVimMode();
 
   const setIsBackgroundTaskListOpenRef = useRef<(open: boolean) => void>(
@@ -969,6 +977,7 @@ Logging in with Google... Restarting Gemini CLI to continue.
       openSettingsDialog,
       openSessionBrowser,
       openModelDialog,
+      openVoiceModelDialog,
       openAgentConfigDialog,
       openPermissionsDialog,
       quit: (messages: HistoryItem[]) => {
@@ -981,6 +990,7 @@ Logging in with Google... Restarting Gemini CLI to continue.
       },
       setDebugMessage,
       toggleCorgiMode: () => setCorgiMode((prev) => !prev),
+      toggleVoiceMode: () => setVoiceModeEnabled((prev) => !prev),
       toggleDebugProfiler,
       dispatchExtensionStateUpdate,
       addConfirmUpdateExtensionRequest,
@@ -1006,6 +1016,7 @@ Logging in with Google... Restarting Gemini CLI to continue.
       openSettingsDialog,
       openSessionBrowser,
       openModelDialog,
+      openVoiceModelDialog,
       openAgentConfigDialog,
       setQuittingMessages,
       setDebugMessage,
@@ -2191,6 +2202,7 @@ Logging in with Google... Restarting Gemini CLI to continue.
     isThemeDialogOpen ||
     isSettingsDialogOpen ||
     isModelDialogOpen ||
+    isVoiceModelDialogOpen ||
     isAgentConfigDialogOpen ||
     isPermissionsDialogOpen ||
     isAuthenticating ||
@@ -2448,6 +2460,7 @@ Logging in with Google... Restarting Gemini CLI to continue.
       isSettingsDialogOpen,
       isSessionBrowserOpen,
       isModelDialogOpen,
+      isVoiceModelDialogOpen,
       isAgentConfigDialogOpen,
       selectedAgentName,
       selectedAgentDisplayName,
@@ -2468,6 +2481,7 @@ Logging in with Google... Restarting Gemini CLI to continue.
       pendingGeminiHistoryItems,
       thought,
       isInputActive,
+      isVoiceModeEnabled,
       isResuming,
       shouldShowIdePrompt,
       isFolderTrustDialogOpen: isFolderTrustDialogOpen ?? false,
@@ -2559,6 +2573,7 @@ Logging in with Google... Restarting Gemini CLI to continue.
       isSettingsDialogOpen,
       isSessionBrowserOpen,
       isModelDialogOpen,
+      isVoiceModelDialogOpen,
       isAgentConfigDialogOpen,
       selectedAgentName,
       selectedAgentDisplayName,
@@ -2579,6 +2594,7 @@ Logging in with Google... Restarting Gemini CLI to continue.
       pendingGeminiHistoryItems,
       thought,
       isInputActive,
+      isVoiceModeEnabled,
       isResuming,
       shouldShowIdePrompt,
       isFolderTrustDialogOpen,
@@ -2671,6 +2687,8 @@ Logging in with Google... Restarting Gemini CLI to continue.
       exitPrivacyNotice,
       closeSettingsDialog,
       closeModelDialog,
+      openVoiceModelDialog,
+      closeVoiceModelDialog,
       openAgentConfigDialog,
       closeAgentConfigDialog,
       openPermissionsDialog,
@@ -2751,6 +2769,9 @@ Logging in with Google... Restarting Gemini CLI to continue.
         setAccountSuspensionInfo(null);
         setAuthState(AuthState.Updating);
       },
+      setVoiceModeEnabled: (value: boolean) => {
+        setVoiceModeEnabled(value);
+      },
     }),
     [
       handleThemeSelect,
@@ -2764,6 +2785,8 @@ Logging in with Google... Restarting Gemini CLI to continue.
       exitPrivacyNotice,
       closeSettingsDialog,
       closeModelDialog,
+      openVoiceModelDialog,
+      closeVoiceModelDialog,
       openAgentConfigDialog,
       closeAgentConfigDialog,
       openPermissionsDialog,
@@ -2807,6 +2830,7 @@ Logging in with Google... Restarting Gemini CLI to continue.
       config,
       historyManager,
       getPreferredEditor,
+      setVoiceModeEnabled,
     ],
   );
 

packages/cli/src/ui/commands/quitCommand.test.ts

@@ -33,11 +33,12 @@ describe('quitCommand', () => {
     });
 
     if (!quitCommand.action) throw new Error('Action is not defined');
-    const result = quitCommand.action(mockContext, 'quit');
+    const result = quitCommand.action(mockContext, '');
 
     expect(formatDuration).toHaveBeenCalledWith(3600000); // 1 hour in ms
     expect(result).toEqual({
       type: 'quit',
+      deleteSession: false,
       messages: [
         {
           type: 'user',
@@ -52,4 +53,54 @@ describe('quitCommand', () => {
       ],
     });
   });
+
+  it('sets deleteSession to true when --delete flag is provided', () => {
+    const mockContext = createMockCommandContext({
+      session: {
+        stats: {
+          sessionStartTime: new Date('2025-01-01T00:00:00Z'),
+        },
+      },
+    });
+
+    if (!quitCommand.action) throw new Error('Action is not defined');
+    const result = quitCommand.action(mockContext, '--delete');
+
+    expect(result).toEqual({
+      type: 'quit',
+      deleteSession: true,
+      messages: [
+        {
+          type: 'user',
+          text: '/quit',
+          id: expect.any(Number),
+        },
+        {
+          type: 'quit',
+          duration: '1h 0m 0s',
+          id: expect.any(Number),
+        },
+      ],
+    });
+  });
+
+  it('does not set deleteSession for unrecognized args', () => {
+    const mockContext = createMockCommandContext({
+      session: {
+        stats: {
+          sessionStartTime: new Date('2025-01-01T00:00:00Z'),
+        },
+      },
+    });
+
+    if (!quitCommand.action) throw new Error('Action is not defined');
+    const result = quitCommand.action(mockContext, 'some-random-arg');
+
+    expect(result).toEqual(
+      expect.objectContaining({
+        type: 'quit',
+        deleteSession: false,
+      }),
+    );
+  });
 });

packages/cli/src/ui/commands/quitCommand.ts

@@ -13,13 +13,16 @@ export const quitCommand: SlashCommand = {
   description: 'Exit the cli',
   kind: CommandKind.BUILT_IN,
   autoExecute: true,
-  action: (context) => {
+  action: (context, args) => {
     const now = Date.now();
     const { sessionStartTime } = context.session.stats;
     const wallDuration = now - sessionStartTime.getTime();
 
+    const deleteSession = args.trim() === '--delete';
+
     return {
       type: 'quit',
+      deleteSession,
       messages: [
         {
           type: 'user',

packages/cli/src/ui/commands/types.ts

@@ -72,6 +72,7 @@ export interface CommandContext {
     loadHistory: (history: HistoryItem[], postLoadInput?: string) => void;
     /** Toggles a special display mode. */
     toggleCorgiMode: () => void;
+    toggleVoiceMode: () => void;
     toggleDebugProfiler: () => void;
     toggleVimEnabled: () => Promise<boolean>;
     reloadCommands: () => void;
@@ -107,6 +108,8 @@ export interface CommandContext {
 export interface QuitActionReturn {
   type: 'quit';
   messages: HistoryItem[];
+  /** When true, the current session's history and temporary files will be deleted on exit. */
+  deleteSession?: boolean;
 }
 
 /**
@@ -125,6 +128,7 @@ export interface OpenDialogActionReturn {
     | 'settings'
     | 'sessionBrowser'
     | 'model'
+    | 'voice-model'
     | 'agentConfig'
     | 'permissions';
 }

packages/cli/src/ui/commands/voiceCommand.ts

@@ -0,0 +1,30 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import { CommandKind, type SlashCommand } from './types.js';
+
+export const voiceCommand: SlashCommand = {
+  name: 'voice',
+  altNames: [],
+  description: 'Toggle voice dictation mode',
+  kind: CommandKind.BUILT_IN,
+  autoExecute: true,
+  action: (context) => {
+    context.ui.toggleVoiceMode();
+  },
+  subCommands: [
+    {
+      name: 'model',
+      description: 'Manage voice transcription models',
+      kind: CommandKind.BUILT_IN,
+      autoExecute: true,
+      action: async () => ({
+        type: 'dialog',
+        dialog: 'voice-model',
+      }),
+    },
+  ],
+};

packages/cli/src/ui/components/DialogManager.tsx

@@ -25,6 +25,7 @@ import { relaunchApp } from '../../utils/processUtils.js';
 import { SessionBrowser } from './SessionBrowser.js';
 import { PermissionsModifyTrustDialog } from './PermissionsModifyTrustDialog.js';
 import { ModelDialog } from './ModelDialog.js';
+import { VoiceModelDialog } from './VoiceModelDialog.js';
 import { theme } from '../semantic-colors.js';
 import { useUIState } from '../contexts/UIStateContext.js';
 import { useQuotaState } from '../contexts/QuotaContext.js';
@@ -238,6 +239,9 @@ export const DialogManager = ({
   if (uiState.isModelDialogOpen) {
     return <ModelDialog onClose={uiActions.closeModelDialog} />;
   }
+  if (uiState.isVoiceModelDialogOpen) {
+    return <VoiceModelDialog onClose={uiActions.closeVoiceModelDialog} />;
+  }
   if (
     uiState.isAgentConfigDialogOpen &&
     uiState.selectedAgentName &&

packages/cli/src/ui/components/InputPrompt.test.tsx

@@ -9,12 +9,41 @@ import { createMockSettings } from '../../test-utils/settings.js';
 import { makeFakeConfig } from '@google/gemini-cli-core';
 import { waitFor } from '../../test-utils/async.js';
 import { act, useState, useMemo } from 'react';
+import type { EventEmitter } from 'node:events';
+
+const { fakeTranscriptionProvider } = vi.hoisted(() => {
+  // Use require within hoisted block for immediate synchronous access
+  // eslint-disable-next-line @typescript-eslint/no-require-imports, no-restricted-syntax
+  const { EventEmitter } = require('node:events');
+  class FakeTranscriptionProvider extends EventEmitter {
+    connect = vi.fn().mockResolvedValue(undefined);
+    disconnect = vi.fn();
+    sendAudioChunk = vi.fn();
+    getTranscription = vi.fn().mockReturnValue('');
+  }
+  return {
+    fakeTranscriptionProvider: new FakeTranscriptionProvider(),
+  };
+});
+
+vi.mock('@google/gemini-cli-core', async (importOriginal) => {
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
+  const actual = (await importOriginal()) as any;
+  return {
+    ...actual,
+    TranscriptionFactory: {
+      createProvider: vi.fn(() => fakeTranscriptionProvider),
+    },
+  };
+});
+
 import {
   InputPrompt,
   tryTogglePasteExpansion,
   type InputPromptProps,
 } from './InputPrompt.js';
 import { InputContext } from '../contexts/InputContext.js';
+import { type UIState } from '../contexts/UIStateContext.js';
 import {
   calculateTransformationsForLine,
   calculateTransformedLine,
@@ -417,6 +446,7 @@ describe('InputPrompt', () => {
         getWorkspaceContext: () => ({
           getDirectories: () => ['/test/project/src'],
         }),
+        getContentGeneratorConfig: () => ({ apiKey: 'test-api-key' }),
       } as unknown as Config,
       slashCommands: mockSlashCommands,
       commandContext: mockCommandContext,
@@ -4925,6 +4955,383 @@ describe('InputPrompt', () => {
       },
     );
   });
+
+  describe('Voice Mode', () => {
+    beforeEach(() => {
+      (
+        fakeTranscriptionProvider as unknown as EventEmitter
+      ).removeAllListeners();
+      vi.clearAllMocks();
+    });
+
+    it('should start recording when space is pressed and voice mode is enabled (toggle)', async () => {
+      await act(async () => {
+        mockBuffer.setText('');
+      });
+      const { stdin, unmount, lastFrame } = await renderWithProviders(
+        <TestInputPrompt {...props} focus={true} buffer={mockBuffer} />,
+        {
+          uiState: { isVoiceModeEnabled: true } as UIState,
+          settings: createMockSettings({
+            experimental: { voice: { activationMode: 'toggle' } },
+          }),
+        },
+      );
+
+      // Initially not recording
+      expect(lastFrame()).not.toContain('🎙️ Listening...');
+      expect(lastFrame()).toContain(
+        'Voice mode: Space to start/stop recording',
+      );
+
+      // Press space to start
+      await act(async () => {
+        stdin.write(' ');
+      });
+
+      // Now should show listening
+      await waitFor(() => {
+        expect(lastFrame()).toContain('🎙️ Listening...');
+      });
+
+      unmount();
+    });
+
+    it('should toggle recording off when space is pressed again (toggle)', async () => {
+      await act(async () => {
+        mockBuffer.setText('');
+      });
+      const { stdin, unmount, lastFrame } = await renderWithProviders(
+        <TestInputPrompt {...props} focus={true} buffer={mockBuffer} />,
+        {
+          uiState: { isVoiceModeEnabled: true } as UIState,
+          settings: createMockSettings({
+            experimental: { voice: { activationMode: 'toggle' } },
+          }),
+        },
+      );
+
+      // Start recording
+      await act(async () => {
+        stdin.write(' ');
+      });
+      await waitFor(() => {
+        expect(lastFrame()).toContain('🎙️ Listening...');
+      });
+
+      // Stop recording
+      await act(async () => {
+        stdin.write(' ');
+      });
+      await waitFor(() => {
+        expect(lastFrame()).not.toContain('🎙️ Listening...');
+        expect(lastFrame()).toContain(
+          'Voice mode: Space to start/stop recording',
+        );
+      });
+
+      unmount();
+    });
+
+    it('should resume recording when space is pressed even if buffer is not empty (toggle)', async () => {
+      await act(async () => {
+        mockBuffer.setText('some existing text');
+      });
+      const { stdin, unmount, lastFrame } = await renderWithProviders(
+        <TestInputPrompt {...props} focus={true} buffer={mockBuffer} />,
+        {
+          uiState: { isVoiceModeEnabled: true } as UIState,
+          settings: createMockSettings({
+            experimental: { voice: { activationMode: 'toggle' } },
+          }),
+        },
+      );
+
+      // Should show voice mode hint even if buffer is not empty (new behavior)
+      expect(lastFrame()).toContain(
+        'Voice mode: Space to start/stop recording',
+      );
+      expect(lastFrame()).toContain('some existing text');
+
+      // Press space to start recording again
+      await act(async () => {
+        stdin.write(' ');
+      });
+
+      await waitFor(() => {
+        expect(lastFrame()).toContain('🎙️ Listening...');
+      });
+
+      unmount();
+    });
+
+    it('should not start recording if voice mode is disabled (toggle)', async () => {
+      await act(async () => {
+        mockBuffer.setText('');
+      });
+      const { stdin, unmount, lastFrame } = await renderWithProviders(
+        <TestInputPrompt {...props} focus={true} buffer={mockBuffer} />,
+        {
+          uiState: { isVoiceModeEnabled: false } as UIState,
+          settings: createMockSettings({
+            experimental: { voice: { activationMode: 'toggle' } },
+          }),
+        },
+      );
+
+      // Press space
+      await act(async () => {
+        stdin.write(' ');
+      });
+
+      // Should NOT show listening, instead should call handleInput which handles space
+      expect(lastFrame()).not.toContain('🎙️ Listening...');
+      expect(mockBuffer.handleInput).toHaveBeenCalled();
+      unmount();
+    });
+
+    it('should append transcription correctly across multiple turn updates (toggle)', async () => {
+      await act(async () => {
+        mockBuffer.setText('initial');
+      });
+      const { stdin, unmount } = await renderWithProviders(
+        <TestInputPrompt {...props} focus={true} buffer={mockBuffer} />,
+        {
+          uiState: { isVoiceModeEnabled: true } as UIState,
+          settings: createMockSettings({
+            experimental: { voice: { activationMode: 'toggle' } },
+          }),
+        },
+      );
+
+      // Start recording
+      await act(async () => {
+        stdin.write(' ');
+      });
+
+      // Emit first transcription
+      await act(async () => {
+        (fakeTranscriptionProvider as unknown as EventEmitter).emit(
+          'transcription',
+          'hello',
+        );
+      });
+      await waitFor(() => {
+        expect(mockBuffer.setText).toHaveBeenCalledWith('initial hello', 'end');
+      });
+
+      // Emit turnComplete (Gemini Live starts over after this)
+      await act(async () => {
+        (fakeTranscriptionProvider as unknown as EventEmitter).emit(
+          'turnComplete',
+        );
+      });
+
+      // Emit second part (Gemini Live sends new turn text starting from empty)
+      await act(async () => {
+        (fakeTranscriptionProvider as unknown as EventEmitter).emit(
+          'transcription',
+          'world',
+        );
+      });
+      await waitFor(() => {
+        // Should have appended 'world' to the baseline 'initial hello'
+        expect(mockBuffer.setText).toHaveBeenCalledWith(
+          'initial hello world',
+          'end',
+        );
+      });
+
+      unmount();
+    });
+
+    it('should append transcription correctly when resuming voice mode (toggle)', async () => {
+      await act(async () => {
+        mockBuffer.setText('First turn.');
+      });
+      const { stdin, unmount } = await renderWithProviders(
+        <TestInputPrompt {...props} focus={true} buffer={mockBuffer} />,
+        {
+          uiState: { isVoiceModeEnabled: true } as UIState,
+          settings: createMockSettings({
+            experimental: { voice: { activationMode: 'toggle' } },
+          }),
+        },
+      );
+
+      // Start recording (resumed)
+      await act(async () => {
+        stdin.write(' ');
+      });
+
+      // Emit transcription
+      await act(async () => {
+        (fakeTranscriptionProvider as unknown as EventEmitter).emit(
+          'transcription',
+          'Second turn.',
+        );
+      });
+
+      await waitFor(() => {
+        expect(mockBuffer.setText).toHaveBeenCalledWith(
+          'First turn. Second turn.',
+          'end',
+        );
+      });
+
+      unmount();
+    });
+
+    describe('push-to-talk', () => {
+      beforeEach(() => {
+        vi.useFakeTimers();
+      });
+
+      afterEach(() => {
+        vi.useRealTimers();
+      });
+
+      it('should insert a space on a single tap', async () => {
+        const { stdin, unmount, lastFrame } = await renderWithProviders(
+          <TestInputPrompt {...props} focus={true} buffer={mockBuffer} />,
+          {
+            uiState: { isVoiceModeEnabled: true } as UIState,
+            settings: createMockSettings({
+              experimental: { voice: { activationMode: 'push-to-talk' } },
+            }),
+          },
+        );
+
+        expect(lastFrame()).toContain('Voice mode: Hold Space to record');
+
+        // Press space once
+        await act(async () => {
+          stdin.write(' ');
+        });
+
+        // Should insert space optimistically
+        expect(mockBuffer.insert).toHaveBeenCalledWith(' ');
+        expect(lastFrame()).not.toContain('🎙️ Listening...');
+
+        // Advance timer past HOLD_DELAY_MS
+        await act(async () => {
+          vi.advanceTimersByTime(700);
+        });
+
+        expect(lastFrame()).not.toContain('🎙️ Listening...');
+        unmount();
+      });
+
+      it('should start recording on hold (simulated by repeat spaces)', async () => {
+        const { stdin, unmount, lastFrame } = await renderWithProviders(
+          <TestInputPrompt {...props} focus={true} buffer={mockBuffer} />,
+          {
+            uiState: { isVoiceModeEnabled: true } as UIState,
+            settings: createMockSettings({
+              experimental: { voice: { activationMode: 'push-to-talk' } },
+            }),
+          },
+        );
+
+        // First space
+        await act(async () => {
+          stdin.write(' ');
+        });
+        expect(mockBuffer.insert).toHaveBeenCalledWith(' ');
+
+        // Second space (repeat)
+        await act(async () => {
+          stdin.write(' ');
+        });
+
+        await waitFor(() => {
+          // Should have backspaced the optimistic space
+          expect(mockBuffer.backspace).toHaveBeenCalled();
+          // Should show listening
+          expect(lastFrame()).toContain('🎙️ Listening...');
+        });
+
+        unmount();
+      });
+
+      it('should stop recording when space heartbeat stops (release)', async () => {
+        const { stdin, unmount, lastFrame } = await renderWithProviders(
+          <TestInputPrompt {...props} focus={true} buffer={mockBuffer} />,
+          {
+            uiState: { isVoiceModeEnabled: true } as UIState,
+            settings: createMockSettings({
+              experimental: { voice: { activationMode: 'push-to-talk' } },
+            }),
+          },
+        );
+
+        // Start hold
+        await act(async () => {
+          stdin.write(' ');
+          stdin.write(' ');
+        });
+
+        // Use a short interval in waitFor to prevent advancing fake timers past the 300ms RELEASE_DELAY_MS
+        await waitFor(
+          () => {
+            expect(lastFrame()).toContain('🎙️ Listening...');
+          },
+          { interval: 10 },
+        );
+
+        // Simulate heartbeat (held key) - send space first to reset timer, then advance
+        await act(async () => {
+          stdin.write(' ');
+          vi.advanceTimersByTime(100);
+        });
+        expect(lastFrame()).toContain('🎙️ Listening...');
+
+        // Stop heartbeat (release)
+        await act(async () => {
+          vi.advanceTimersByTime(400); // Past RELEASE_DELAY_MS
+        });
+
+        await waitFor(() => {
+          expect(lastFrame()).not.toContain('🎙️ Listening...');
+        });
+
+        unmount();
+      });
+
+      it('should cancel hold state if non-space key is pressed after first space', async () => {
+        const { stdin, unmount } = await renderWithProviders(
+          <TestInputPrompt {...props} focus={true} buffer={mockBuffer} />,
+          {
+            uiState: { isVoiceModeEnabled: true } as UIState,
+            settings: createMockSettings({
+              experimental: { voice: { activationMode: 'push-to-talk' } },
+            }),
+          },
+        );
+
+        // First space
+        await act(async () => {
+          stdin.write(' ');
+        });
+
+        // Type 'a'
+        await act(async () => {
+          stdin.write('a');
+        });
+
+        // Should NOT start recording on next space even if fast
+        await act(async () => {
+          stdin.write(' ');
+        });
+
+        expect(mockBuffer.insert).toHaveBeenCalledTimes(2); // Two spaces inserted
+        expect(mockBuffer.handleInput).toHaveBeenCalledWith(
+          expect.objectContaining({ name: 'a' }),
+        );
+        unmount();
+      });
+    });
+  });
 });
 
 function clean(str: string | undefined): string {

packages/cli/src/ui/components/InputPrompt.tsx

@@ -56,6 +56,7 @@ import {
   debugLogger,
   type Config,
 } from '@google/gemini-cli-core';
+import { useVoiceMode } from '../hooks/useVoiceMode.js';
 import {
   parseInputForHighlighting,
   parseSegmentsFromTokens,
@@ -159,7 +160,6 @@ export function isLargePaste(text: string): boolean {
 }
 
 const DOUBLE_TAB_CLEAN_UI_TOGGLE_WINDOW_MS = 350;
-
 /**
  * Attempt to toggle expansion of a paste placeholder in the buffer.
  * Returns true if a toggle action was performed or hint was shown, false otherwise.
@@ -238,6 +238,7 @@ export const InputPrompt: React.FC<InputPromptProps> = ({
     setEmbeddedShellFocused,
     setShortcutsHelpVisible,
     toggleCleanUiDetailsVisible,
+    setVoiceModeEnabled,
   } = useUIActions();
   const {
     terminalWidth,
@@ -246,6 +247,7 @@ export const InputPrompt: React.FC<InputPromptProps> = ({
     backgroundTasks,
     backgroundTaskHeight,
     shortcutsHelpVisible,
+    isVoiceModeEnabled,
   } = useUIState();
   const [suppressCompletion, setSuppressCompletion] = useState(false);
   const { handlePress: registerPlainTabPress, resetCount: resetPlainTabPress } =
@@ -263,6 +265,7 @@ export const InputPrompt: React.FC<InputPromptProps> = ({
           resetEscapeState();
           if (buffer.text.length > 0) {
             buffer.setText('');
+            resetTurnBaseline();
             resetCompletionState();
           } else if (history.length > 0) {
             onSubmit('/rewind');
@@ -281,6 +284,16 @@ export const InputPrompt: React.FC<InputPromptProps> = ({
   const hasUserNavigatedSuggestions = useRef(false);
   const listRef = useRef<ScrollableListRef<ScrollableItem>>(null);
 
+  const { isRecording, handleVoiceInput, resetTurnBaseline } = useVoiceMode({
+    buffer,
+    config,
+    settings,
+    setQueueErrorMessage,
+    isVoiceModeEnabled,
+    setVoiceModeEnabled,
+    keyMatchers,
+  });
+
   const [reverseSearchActive, setReverseSearchActive] = useState(false);
   const [commandSearchActive, setCommandSearchActive] = useState(false);
   const [textBeforeReverseSearch, setTextBeforeReverseSearch] = useState('');
@@ -387,6 +400,7 @@ export const InputPrompt: React.FC<InputPromptProps> = ({
       // Clear the buffer *before* calling onSubmit to prevent potential re-submission
       // if onSubmit triggers a re-render while the buffer still holds the old value.
       buffer.setText('');
+      resetTurnBaseline();
       onSubmit(processedValue);
       resetCompletionState();
       resetReverseSearchCompletionState();
@@ -398,6 +412,7 @@ export const InputPrompt: React.FC<InputPromptProps> = ({
       shellModeActive,
       shellHistory,
       resetReverseSearchCompletionState,
+      resetTurnBaseline,
     ],
   );
 
@@ -647,6 +662,8 @@ export const InputPrompt: React.FC<InputPromptProps> = ({
 
   const handleInput = useCallback(
     (key: Key) => {
+      if (handleVoiceInput(key)) return true;
+
       // Determine if this keypress is a history navigation command
       const isHistoryUp =
         !shellModeActive &&
@@ -873,9 +890,9 @@ export const InputPrompt: React.FC<InputPromptProps> = ({
       ) {
         setShellModeActive(!shellModeActive);
         buffer.setText(''); // Clear the '!' from input
+        resetTurnBaseline();
         return true;
       }
-
       if (keyMatchers[Command.ESCAPE](key)) {
         const cancelSearch = (
           setActive: (active: boolean) => void,
@@ -1360,6 +1377,7 @@ export const InputPrompt: React.FC<InputPromptProps> = ({
       backgroundTaskHeight,
       streamingState,
       handleEscPress,
+      resetTurnBaseline,
       registerPlainTabPress,
       resetPlainTabPress,
       toggleCleanUiDetailsVisible,
@@ -1369,9 +1387,9 @@ export const InputPrompt: React.FC<InputPromptProps> = ({
       keyMatchers,
       isHelpDismissKey,
       settings,
+      handleVoiceInput,
     ],
   );
-
   useKeypress(handleInput, {
     isActive: !isEmbeddedShellFocused && !copyModeEnabled,
     priority: true,
@@ -1792,20 +1810,39 @@ export const InputPrompt: React.FC<InputPromptProps> = ({
             )}{' '}
           </Text>
           <Box flexGrow={1} flexDirection="column" ref={innerBoxRef}>
-            {buffer.text.length === 0 && placeholder ? (
-              showCursor ? (
-                <Text
-                  terminalCursorFocus={showCursor}
-                  terminalCursorPosition={0}
-                >
-                  {chalk.inverse(placeholder.slice(0, 1))}
-                  <Text color={theme.text.secondary}>
-                    {placeholder.slice(1)}
-                  </Text>
+            {isRecording && (
+              <Box flexDirection="row" marginBottom={0}>
+                <Text color={theme.status.success}>🎙️ Listening...</Text>
+              </Box>
+            )}
+            {isVoiceModeEnabled && !isRecording && (
+              <Box flexDirection="row" marginBottom={0}>
+                <Text color={theme.text.secondary}>
+                  &gt; Voice mode:{' '}
+                  {(settings.experimental.voice?.activationMode ??
+                    'push-to-talk') === 'push-to-talk'
+                    ? 'Hold Space to record'
+                    : 'Space to start/stop recording'}{' '}
+                  (Esc to exit)
                 </Text>
-              ) : (
-                <Text color={theme.text.secondary}>{placeholder}</Text>
-              )
+              </Box>
+            )}
+            {buffer.text.length === 0 && !isRecording ? (
+              !isVoiceModeEnabled && placeholder ? (
+                showCursor ? (
+                  <Text
+                    terminalCursorFocus={showCursor}
+                    terminalCursorPosition={0}
+                  >
+                    {chalk.inverse(placeholder.slice(0, 1))}
+                    <Text color={theme.text.secondary}>
+                      {placeholder.slice(1)}
+                    </Text>
+                  </Text>
+                ) : (
+                  <Text color={theme.text.secondary}>{placeholder}</Text>
+                )
+              ) : null
             ) : (
               <Box
                 flexDirection="column"

packages/cli/src/ui/components/SettingsDialog.test.tsx

@@ -44,7 +44,7 @@ enum TerminalKeys {
   LEFT_ARROW = '\u001B[D',
   RIGHT_ARROW = '\u001B[C',
   ESCAPE = '\u001B',
-  BACKSPACE = '\x7f',
+  BACKSPACE = '\u0008',
   CTRL_P = '\u0010',
   CTRL_N = '\u000E',
 }

packages/cli/src/ui/components/VoiceModelDialog.tsx

@@ -0,0 +1,236 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type React from 'react';
+import { useCallback, useMemo, useState } from 'react';
+import { Box, Text } from 'ink';
+import { theme } from '../semantic-colors.js';
+import { DescriptiveRadioButtonSelect } from './shared/DescriptiveRadioButtonSelect.js';
+import { useSettingsStore } from '../contexts/SettingsContext.js';
+import { SettingScope } from '../../config/settings.js';
+import { useKeypress, type Key } from '../hooks/useKeypress.js';
+import { isBinaryAvailable } from '@google/gemini-cli-core';
+import {
+  WhisperModelManager,
+  type WhisperModelProgress,
+} from '@google/gemini-cli-core';
+import { CliSpinner } from './CliSpinner.js';
+
+interface VoiceModelDialogProps {
+  onClose: () => void;
+}
+
+type DialogView = 'backend' | 'whisper-models';
+
+const WHISPER_MODELS = [
+  {
+    value: 'ggml-tiny.en.bin',
+    label: 'Tiny (EN)',
+    description: 'Fastest, lower accuracy (~75MB)',
+  },
+  {
+    value: 'ggml-base.en.bin',
+    label: 'Base (EN)',
+    description: 'Balanced speed and accuracy (~142MB)',
+  },
+  {
+    value: 'ggml-large-v3-turbo-q5_0.bin',
+    label: 'Large v3 Turbo (Q5_0)',
+    description: 'High accuracy, quantized (~547MB)',
+  },
+  {
+    value: 'ggml-large-v3-turbo-q8_0.bin',
+    label: 'Large v3 Turbo (Q8_0)',
+    description: 'Maximum accuracy, high memory (~834MB)',
+  },
+];
+
+export function VoiceModelDialog({
+  onClose,
+}: VoiceModelDialogProps): React.JSX.Element {
+  const { settings, setSetting } = useSettingsStore();
+  const [view, setView] = useState<DialogView>('backend');
+  const [downloadProgress, setDownloadProgress] =
+    useState<WhisperModelProgress | null>(null);
+  const [error, setError] = useState<string | null>(null);
+
+  const whisperInstalled = useMemo(
+    () => isBinaryAvailable('whisper-stream'),
+    [],
+  );
+  const modelManager = useMemo(() => new WhisperModelManager(), []);
+
+  const currentBackend =
+    settings.merged.experimental.voice?.backend ?? 'gemini-live';
+  const currentWhisperModel =
+    settings.merged.experimental.voice?.whisperModel ?? 'ggml-base.en.bin';
+
+  const handleKeypress = useCallback(
+    (key: Key) => {
+      if (key.name === 'escape') {
+        if (view === 'whisper-models') {
+          setView('backend');
+        } else {
+          onClose();
+        }
+        return true;
+      }
+      return false;
+    },
+    [view, onClose],
+  );
+
+  useKeypress(handleKeypress, { isActive: true });
+
+  const handleBackendSelect = useCallback(
+    (value: string) => {
+      if (value === 'whisper') {
+        setView('whisper-models');
+      } else {
+        setSetting(
+          SettingScope.User,
+          'experimental.voice.backend',
+          'gemini-live',
+        );
+        onClose();
+      }
+    },
+    [setSetting, onClose],
+  );
+
+  const handleWhisperModelSelect = useCallback(
+    async (modelName: string) => {
+      if (modelManager.isModelInstalled(modelName)) {
+        setSetting(SettingScope.User, 'experimental.voice.backend', 'whisper');
+        setSetting(
+          SettingScope.User,
+          'experimental.voice.whisperModel',
+          modelName,
+        );
+        onClose();
+      } else {
+        setError(null);
+        const onProgress = (p: WhisperModelProgress) => setDownloadProgress(p);
+        modelManager.on('progress', onProgress);
+
+        try {
+          await modelManager.downloadModel(modelName);
+
+          setSetting(
+            SettingScope.User,
+            'experimental.voice.backend',
+            'whisper',
+          );
+          setSetting(
+            SettingScope.User,
+            'experimental.voice.whisperModel',
+            modelName,
+          );
+          onClose();
+        } catch (err) {
+          setError(
+            `Failed to download: ${err instanceof Error ? err.message : String(err)}`,
+          );
+        } finally {
+          modelManager.off('progress', onProgress);
+          setDownloadProgress(null);
+        }
+      }
+    },
+    [modelManager, setSetting, onClose],
+  );
+
+  const backendOptions = useMemo(
+    () => [
+      {
+        value: 'gemini-live',
+        title: 'Gemini Live API (Cloud)',
+        description: 'Real-time cloud transcription via Gemini Live API.',
+        key: 'gemini-live',
+      },
+      {
+        value: 'whisper',
+        title: 'Whisper (Local)',
+        description: whisperInstalled
+          ? 'Local transcription using whisper.cpp.'
+          : 'Local transcription (Requires: brew install whisper-cpp)',
+        key: 'whisper',
+      },
+    ],
+    [whisperInstalled],
+  );
+
+  const whisperOptions = useMemo(
+    () =>
+      WHISPER_MODELS.map((m) => ({
+        value: m.value,
+        title: `${m.label}${modelManager.isModelInstalled(m.value) ? ' (Installed)' : ' (Download)'}`,
+        description: m.description,
+        key: m.value,
+      })),
+    [modelManager],
+  );
+
+  return (
+    <Box
+      borderStyle="round"
+      borderColor={theme.border.default}
+      flexDirection="column"
+      padding={1}
+      width="100%"
+    >
+      <Text bold>
+        {view === 'backend'
+          ? 'Select Voice Transcription Backend'
+          : 'Select Whisper Model'}
+      </Text>
+
+      {error && (
+        <Box marginTop={1}>
+          <Text color={theme.status.error}>{error}</Text>
+        </Box>
+      )}
+
+      {downloadProgress ? (
+        <Box marginTop={1} flexDirection="column">
+          <Box>
+            <Text>Downloading {downloadProgress.modelName}... </Text>
+            <CliSpinner />
+            <Text> {Math.round(downloadProgress.percentage * 100)}%</Text>
+          </Box>
+        </Box>
+      ) : (
+        <Box marginTop={1}>
+          {view === 'backend' ? (
+            <DescriptiveRadioButtonSelect
+              items={backendOptions}
+              onSelect={handleBackendSelect}
+              initialIndex={currentBackend === 'whisper' ? 1 : 0}
+              showNumbers={true}
+            />
+          ) : (
+            <DescriptiveRadioButtonSelect
+              items={whisperOptions}
+              onSelect={handleWhisperModelSelect}
+              initialIndex={whisperOptions.findIndex(
+                (o) => o.value === currentWhisperModel,
+              )}
+              showNumbers={true}
+            />
+          )}
+        </Box>
+      )}
+
+      <Box marginTop={1} flexDirection="column">
+        <Text color={theme.text.secondary}>
+          {view === 'whisper-models'
+            ? '(Press Esc to go back)'
+            : '(Press Esc to close)'}
+        </Text>
+      </Box>
+    </Box>
+  );
+}

packages/cli/src/ui/components/__snapshots__/InputPrompt.test.tsx.snap

@@ -168,13 +168,6 @@ exports[`InputPrompt > mouse interaction > should toggle paste expansion on doub
 "
 `;
 
-exports[`InputPrompt > mouse interaction > should toggle paste expansion on double-click 4`] = `
-"▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀
- > [Pasted Text: 10 lines]                                                                          
-▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄
-"
-`;
-
 exports[`InputPrompt > multiline rendering > should correctly render multiline input including blank lines 1`] = `
 "────────────────────────────────────────────────────────────────────────────────────────────────────
  > hello

packages/cli/src/ui/components/shared/BaseSettingsDialog.test.tsx

@@ -24,7 +24,7 @@ enum TerminalKeys {
   LEFT_ARROW = '\u001B[D',
   RIGHT_ARROW = '\u001B[C',
   ESCAPE = '\u001B',
-  BACKSPACE = '\x7f',
+  BACKSPACE = '\u0008',
   CTRL_L = '\u000C',
 }
 

packages/cli/src/ui/contexts/KeypressContext.test.tsx

@@ -9,17 +9,7 @@ import { act } from 'react';
 import { renderHookWithProviders } from '../../test-utils/render.js';
 import { createMockSettings } from '../../test-utils/settings.js';
 import { waitFor } from '../../test-utils/async.js';
-import type { Mock } from 'vitest';
-import {
-  vi,
-  afterAll,
-  beforeAll,
-  describe,
-  it,
-  expect,
-  beforeEach,
-  afterEach,
-} from 'vitest';
+import { vi, afterAll, beforeAll, type Mock } from 'vitest';
 import {
   useKeypressContext,
   ESC_TIMEOUT,
@@ -441,80 +431,6 @@ describe('KeypressContext', () => {
     );
   });
 
-  describe('Windows Terminal Backspace handling', () => {
-    afterEach(() => {
-      vi.unstubAllEnvs();
-    });
-
-    it('should NOT treat \\b as ctrl when WT_SESSION is NOT present and OS is not Windows_NT', async () => {
-      vi.stubEnv('WT_SESSION', '');
-      vi.stubEnv('OS', 'Linux');
-      const { keyHandler } = await setupKeypressTest();
-
-      act(() => {
-        stdin.write('\b');
-      });
-
-      expect(keyHandler).toHaveBeenCalledWith(
-        expect.objectContaining({
-          name: 'backspace',
-          ctrl: false,
-        }),
-      );
-    });
-
-    it('should treat \\b as ctrl when WT_SESSION IS present (even if not Windows_NT)', async () => {
-      vi.stubEnv('WT_SESSION', 'some-id');
-      vi.stubEnv('OS', 'Linux');
-      const { keyHandler } = await setupKeypressTest();
-
-      act(() => {
-        stdin.write('\b');
-      });
-
-      expect(keyHandler).toHaveBeenCalledWith(
-        expect.objectContaining({
-          name: 'backspace',
-          ctrl: true,
-        }),
-      );
-    });
-
-    it('should treat \\b as ctrl when OS is Windows_NT', async () => {
-      vi.stubEnv('WT_SESSION', '');
-      vi.stubEnv('OS', 'Windows_NT');
-      const { keyHandler } = await setupKeypressTest();
-
-      act(() => {
-        stdin.write('\b');
-      });
-
-      expect(keyHandler).toHaveBeenCalledWith(
-        expect.objectContaining({
-          name: 'backspace',
-          ctrl: true,
-        }),
-      );
-    });
-
-    it('should treat \\x7f as regular backspace regardless of WT_SESSION or OS', async () => {
-      vi.stubEnv('WT_SESSION', 'some-id');
-      vi.stubEnv('OS', 'Windows_NT');
-      const { keyHandler } = await setupKeypressTest();
-
-      act(() => {
-        stdin.write('\x7f');
-      });
-
-      expect(keyHandler).toHaveBeenCalledWith(
-        expect.objectContaining({
-          name: 'backspace',
-          ctrl: false,
-        }),
-      );
-    });
-  });
-
   describe('paste mode', () => {
     it.each([
       {

packages/cli/src/ui/contexts/KeypressContext.tsx

@@ -41,6 +41,7 @@ const KEY_INFO_MAP: Record<
   string,
   { name: string; shift?: boolean; ctrl?: boolean }
 > = {
+  OM: { name: 'enter' },
   '[200~': { name: 'paste-start' },
   '[201~': { name: 'paste-end' },
   '[[A': { name: 'f1' },
@@ -651,20 +652,8 @@ function* emitKeys(
       // tab
       name = 'tab';
       alt = escaped;
-    } else if (ch === '\b') {
-      // ctrl+h / ctrl+backspace (windows terminals send \x08 for ctrl+backspace)
-      name = 'backspace';
-      // In Windows environments, \b is sent for Ctrl+Backspace (standard backspace is translated to \x7f).
-      // We scope this to Windows/WT_SESSION to avoid breaking other unixes where \b is a plain backspace.
-      if (
-        typeof process !== 'undefined' &&
-        (process.env?.['OS'] === 'Windows_NT' || !!process.env?.['WT_SESSION'])
-      ) {
-        ctrl = true;
-      }
-      alt = escaped;
-    } else if (ch === '\x7f') {
-      // backspace
+    } else if (ch === '\b' || ch === '\x7f') {
+      // backspace or ctrl+h
       name = 'backspace';
       alt = escaped;
     } else if (ch === ESC) {

packages/cli/src/ui/contexts/UIActionsContext.tsx

@@ -41,6 +41,8 @@ export interface UIActions {
   exitPrivacyNotice: () => void;
   closeSettingsDialog: () => void;
   closeModelDialog: () => void;
+  openVoiceModelDialog: () => void;
+  closeVoiceModelDialog: () => void;
   openAgentConfigDialog: (
     name: string,
     displayName: string,
@@ -93,6 +95,7 @@ export interface UIActions {
   handleNewAgentsSelect: (choice: NewAgentsChoice) => Promise<void>;
   getPreferredEditor: () => EditorType | undefined;
   clearAccountSuspension: () => void;
+  setVoiceModeEnabled: (value: boolean) => void;
 }
 
 export const UIActionsContext = createContext<UIActions | null>(null);

packages/cli/src/ui/contexts/UIStateContext.tsx

@@ -112,6 +112,7 @@ export interface UIState {
   isSettingsDialogOpen: boolean;
   isSessionBrowserOpen: boolean;
   isModelDialogOpen: boolean;
+  isVoiceModelDialogOpen: boolean;
   isAgentConfigDialogOpen: boolean;
   selectedAgentName?: string;
   selectedAgentDisplayName?: string;
@@ -132,6 +133,7 @@ export interface UIState {
   pendingGeminiHistoryItems: HistoryItemWithoutId[];
   thought: ThoughtSummary | null;
   isInputActive: boolean;
+  isVoiceModeEnabled: boolean;
   isResuming: boolean;
   shouldShowIdePrompt: boolean;
   isFolderTrustDialogOpen: boolean;

packages/cli/src/ui/hooks/slashCommandProcessor.test.tsx

@@ -205,11 +205,13 @@ describe('useSlashCommandProcessor', () => {
             openSettingsDialog: vi.fn(),
             openSessionBrowser: vi.fn(),
             openModelDialog: mockOpenModelDialog,
+            openVoiceModelDialog: vi.fn(),
             openAgentConfigDialog,
             openPermissionsDialog: vi.fn(),
             quit: mockSetQuittingMessages,
             setDebugMessage: vi.fn(),
             toggleCorgiMode: vi.fn(),
+            toggleVoiceMode: vi.fn(),
             toggleDebugProfiler: vi.fn(),
             dispatchExtensionStateUpdate: vi.fn(),
             addConfirmUpdateExtensionRequest: vi.fn(),
@@ -644,6 +646,108 @@ describe('useSlashCommandProcessor', () => {
 
       expect(mockSetQuittingMessages).toHaveBeenCalledWith(['bye']);
     });
+
+    it('should delete the current session when quit action has deleteSession flag', async () => {
+      const mockDeleteCurrentSessionAsync = vi
+        .fn()
+        .mockResolvedValue(undefined);
+
+      const mockClient = {
+        getChatRecordingService: vi.fn().mockReturnValue({
+          deleteCurrentSessionAsync: mockDeleteCurrentSessionAsync,
+        }),
+      } as unknown as GeminiClient;
+      vi.spyOn(mockConfig, 'getGeminiClient').mockReturnValue(mockClient);
+
+      const quitAction = vi.fn().mockResolvedValue({
+        type: 'quit',
+        deleteSession: true,
+        messages: ['bye'],
+      });
+      const command = createTestCommand({
+        name: 'exit',
+        action: quitAction,
+      });
+      const result = await setupProcessorHook({
+        builtinCommands: [command],
+      });
+
+      await waitFor(() => expect(result.current.slashCommands).toHaveLength(1));
+
+      await act(async () => {
+        await result.current.handleSlashCommand('/exit --delete');
+      });
+
+      expect(mockDeleteCurrentSessionAsync).toHaveBeenCalled();
+      expect(mockSetQuittingMessages).toHaveBeenCalledWith(['bye']);
+    });
+
+    it('should not delete session when quit action does not have deleteSession flag', async () => {
+      const mockDeleteCurrentSessionAsync = vi
+        .fn()
+        .mockResolvedValue(undefined);
+      const mockClient = {
+        getChatRecordingService: vi.fn().mockReturnValue({
+          deleteCurrentSessionAsync: mockDeleteCurrentSessionAsync,
+        }),
+      } as unknown as GeminiClient;
+      vi.spyOn(mockConfig, 'getGeminiClient').mockReturnValue(mockClient);
+
+      const quitAction = vi.fn().mockResolvedValue({
+        type: 'quit',
+        messages: ['bye'],
+      });
+      const command = createTestCommand({
+        name: 'exit',
+        action: quitAction,
+      });
+      const result = await setupProcessorHook({
+        builtinCommands: [command],
+      });
+
+      await waitFor(() => expect(result.current.slashCommands).toHaveLength(1));
+
+      await act(async () => {
+        await result.current.handleSlashCommand('/exit');
+      });
+
+      expect(mockDeleteCurrentSessionAsync).not.toHaveBeenCalled();
+      expect(mockSetQuittingMessages).toHaveBeenCalledWith(['bye']);
+    });
+
+    it('should still quit even if session deletion fails', async () => {
+      const mockClient = {
+        getChatRecordingService: vi.fn().mockReturnValue({
+          deleteCurrentSessionAsync: vi
+            .fn()
+            .mockRejectedValue(new Error('Deletion failed')),
+        }),
+      } as unknown as GeminiClient;
+      vi.spyOn(mockConfig, 'getGeminiClient').mockReturnValue(mockClient);
+
+      const quitAction = vi.fn().mockResolvedValue({
+        type: 'quit',
+        deleteSession: true,
+        messages: ['bye'],
+      });
+      const command = createTestCommand({
+        name: 'exit',
+        action: quitAction,
+      });
+      const result = await setupProcessorHook({
+        builtinCommands: [command],
+      });
+
+      await waitFor(() => expect(result.current.slashCommands).toHaveLength(1));
+
+      await act(async () => {
+        await result.current.handleSlashCommand('/exit --delete');
+      });
+
+      // Should still quit even though deletion threw
+      expect(mockSetQuittingMessages).toHaveBeenCalledWith(['bye']);
+    });
+
     it('should handle "submit_prompt" action returned from a file-based command', async () => {
       const fileCommand = createTestCommand(
         {

packages/cli/src/ui/hooks/slashCommandProcessor.ts

@@ -72,6 +72,7 @@ interface SlashCommandProcessorActions {
   openSettingsDialog: () => void;
   openSessionBrowser: () => void;
   openModelDialog: () => void;
+  openVoiceModelDialog: () => void;
   openAgentConfigDialog: (
     name: string,
     displayName: string,
@@ -81,6 +82,7 @@ interface SlashCommandProcessorActions {
   quit: (messages: HistoryItem[]) => void;
   setDebugMessage: (message: string) => void;
   toggleCorgiMode: () => void;
+  toggleVoiceMode: () => void;
   toggleDebugProfiler: () => void;
   dispatchExtensionStateUpdate: (action: ExtensionUpdateAction) => void;
   addConfirmUpdateExtensionRequest: (request: ConfirmationRequest) => void;
@@ -232,6 +234,7 @@ export const useSlashCommandProcessor = (
         pendingItem,
         setPendingItem,
         toggleCorgiMode: actions.toggleCorgiMode,
+        toggleVoiceMode: actions.toggleVoiceMode,
         toggleDebugProfiler: actions.toggleDebugProfiler,
         toggleVimEnabled,
         reloadCommands,
@@ -503,6 +506,9 @@ export const useSlashCommandProcessor = (
                     case 'model':
                       actions.openModelDialog();
                       return { type: 'handled' };
+                    case 'voice-model':
+                      actions.openVoiceModelDialog();
+                      return { type: 'handled' };
                     case 'agentConfig': {
                       // eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
                       const props = result.props as Record<string, unknown>;
@@ -551,6 +557,18 @@ export const useSlashCommandProcessor = (
                   return { type: 'handled' };
                 }
                 case 'quit':
+                  if (result.deleteSession) {
+                    try {
+                      const chatRecordingService = config
+                        ?.getGeminiClient()
+                        ?.getChatRecordingService();
+                      if (chatRecordingService) {
+                        await chatRecordingService.deleteCurrentSessionAsync();
+                      }
+                    } catch {
+                      // Don't let deletion errors prevent exit.
+                    }
+                  }
                   actions.quit(result.messages);
                   return { type: 'handled' };
 

packages/cli/src/ui/hooks/useVoiceMode.ts

@@ -0,0 +1,429 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import { useState, useRef, useCallback, useEffect } from 'react';
+import {
+  AudioRecorder,
+  TranscriptionFactory,
+  debugLogger,
+  type Config,
+  type TranscriptionProvider,
+} from '@google/gemini-cli-core';
+import type { TextBuffer } from '../components/shared/text-buffer.js';
+import type { MergedSettings } from '../../config/settingsSchema.js';
+import type { Key } from './useKeypress.js';
+import { Command } from '../key/keyMatchers.js';
+
+interface UseVoiceModeProps {
+  buffer: TextBuffer;
+  config: Config;
+  settings: MergedSettings;
+  setQueueErrorMessage: (message: string | null) => void;
+  isVoiceModeEnabled: boolean;
+  setVoiceModeEnabled: (enabled: boolean) => void;
+  keyMatchers: Record<Command, (key: Key) => boolean>;
+}
+
+const HOLD_DELAY_MS = 600;
+const RELEASE_DELAY_MS = 300;
+
+export function useVoiceMode({
+  buffer,
+  config,
+  settings,
+  setQueueErrorMessage,
+  isVoiceModeEnabled,
+  setVoiceModeEnabled,
+  keyMatchers,
+}: UseVoiceModeProps) {
+  const [isRecording, setIsRecording] = useState(false);
+  const [isConnecting, setIsConnecting] = useState(false);
+
+  const liveTranscriptionRef = useRef('');
+  const stopRequestedRef = useRef(false);
+  const isRecordingRef = useRef(false);
+  const lastFailureTimeRef = useRef(0);
+  const recordingInProgressRef = useRef(false);
+  const voiceTimeoutRef = useRef<NodeJS.Timeout | null>(null);
+  const recorderRef = useRef<AudioRecorder | null>(null);
+  const transcriptionServiceRef = useRef<TranscriptionProvider | null>(null);
+  const turnBaselineRef = useRef<string | null>(null);
+
+  const pttStateRef = useRef<'idle' | 'possible-hold' | 'recording'>('idle');
+  const pttTimerRef = useRef<NodeJS.Timeout | null>(null);
+  const disconnectTimerRef = useRef<NodeJS.Timeout | null>(null);
+
+  const bufferRef = useRef(buffer);
+  bufferRef.current = buffer;
+
+  const stopVoiceRecording = useCallback(() => {
+    if (stopRequestedRef.current) return;
+    debugLogger.debug('[Voice] Stop requested');
+    stopRequestedRef.current = true;
+
+    setIsRecording(false);
+    isRecordingRef.current = false;
+    setIsConnecting(false);
+
+    if (recorderRef.current) {
+      recorderRef.current.stop();
+      recorderRef.current = null;
+    }
+
+    const serviceToDisconnect = transcriptionServiceRef.current;
+    transcriptionServiceRef.current = null;
+
+    if (serviceToDisconnect) {
+      const isLive = settings.experimental.voice?.backend === 'gemini-live';
+      const gracePeriodMs =
+        settings.experimental.voice?.stopGracePeriodMs ??
+        (isLive ? 2000 : 1000);
+      debugLogger.debug(
+        `[Voice] Draining transcription for ${gracePeriodMs}ms`,
+      );
+
+      if (disconnectTimerRef.current) clearTimeout(disconnectTimerRef.current);
+      disconnectTimerRef.current = setTimeout(() => {
+        debugLogger.debug('[Voice] Grace period ended, disconnecting service');
+        serviceToDisconnect.disconnect();
+        disconnectTimerRef.current = null;
+      }, gracePeriodMs);
+    }
+
+    liveTranscriptionRef.current = '';
+    pttStateRef.current = 'idle';
+  }, [settings.experimental.voice]);
+
+  const startVoiceRecording = useCallback(() => {
+    if (
+      isRecordingRef.current ||
+      Date.now() - lastFailureTimeRef.current < 2000
+    ) {
+      return;
+    }
+
+    if (disconnectTimerRef.current) {
+      clearTimeout(disconnectTimerRef.current);
+      disconnectTimerRef.current = null;
+    }
+
+    recordingInProgressRef.current = true;
+    turnBaselineRef.current = bufferRef.current.text;
+
+    setIsConnecting(true);
+    setIsRecording(true);
+    isRecordingRef.current = true;
+
+    liveTranscriptionRef.current = '';
+    stopRequestedRef.current = false;
+
+    const apiKey =
+      config.getContentGeneratorConfig()?.apiKey ||
+      process.env['GEMINI_API_KEY'] ||
+      '';
+
+    const startAsync = async () => {
+      // If there's an active draining service, disconnect it immediately
+      // before starting a new one to prevent orphaned event collisions.
+      if (disconnectTimerRef.current) {
+        clearTimeout(disconnectTimerRef.current);
+        disconnectTimerRef.current = null;
+      }
+      if (transcriptionServiceRef.current) {
+        transcriptionServiceRef.current.disconnect();
+        transcriptionServiceRef.current = null;
+      }
+
+      const cleanupIfStopped = () => {
+        if (stopRequestedRef.current) {
+          if (recorderRef.current) {
+            recorderRef.current.stop();
+            recorderRef.current = null;
+          }
+          if (transcriptionServiceRef.current) {
+            transcriptionServiceRef.current.disconnect();
+            transcriptionServiceRef.current = null;
+          }
+          setIsRecording(false);
+          isRecordingRef.current = false;
+          setIsConnecting(false);
+          recordingInProgressRef.current = false;
+          return true;
+        }
+        return false;
+      };
+
+      if (cleanupIfStopped()) return;
+
+      const voiceBackend =
+        settings.experimental.voice?.backend ?? 'gemini-live';
+
+      if (!apiKey && voiceBackend === 'gemini-live') {
+        setQueueErrorMessage(
+          'Cloud voice mode requires a GEMINI_API_KEY. Please set it in your environment or ~/.gemini/.env.',
+        );
+        setIsRecording(false);
+        isRecordingRef.current = false;
+        setIsConnecting(false);
+        recordingInProgressRef.current = false;
+        lastFailureTimeRef.current = Date.now();
+        return;
+      }
+
+      if (voiceBackend === 'gemini-live') {
+        recorderRef.current = new AudioRecorder();
+      }
+
+      const currentService = TranscriptionFactory.createProvider(
+        settings.experimental.voice,
+        apiKey,
+      );
+      transcriptionServiceRef.current = currentService;
+
+      currentService.on('transcription', (text) => {
+        if (
+          transcriptionServiceRef.current !== currentService &&
+          stopRequestedRef.current
+        ) {
+          // If this is an orphaned service that was replaced by a new session, ignore its events
+          return;
+        }
+
+        if (text) {
+          const currentBufferText = bufferRef.current.text;
+          const previousTranscription = liveTranscriptionRef.current;
+
+          let newTotalText = currentBufferText;
+
+          if (
+            previousTranscription &&
+            currentBufferText.endsWith(previousTranscription)
+          ) {
+            newTotalText = currentBufferText.slice(
+              0,
+              -previousTranscription.length,
+            );
+          } else if (
+            currentBufferText &&
+            !currentBufferText.endsWith(' ') &&
+            !currentBufferText.endsWith('\n')
+          ) {
+            newTotalText += ' ';
+          }
+
+          newTotalText += text;
+          bufferRef.current.setText(newTotalText, 'end');
+        }
+        liveTranscriptionRef.current = text;
+      });
+
+      currentService.on('turnComplete', () => {
+        if (
+          transcriptionServiceRef.current !== currentService &&
+          stopRequestedRef.current
+        )
+          return;
+        liveTranscriptionRef.current = '';
+      });
+
+      currentService.on('error', (err) => {
+        if (transcriptionServiceRef.current !== currentService) return;
+        debugLogger.error('[Voice] Transcription error:', err);
+        lastFailureTimeRef.current = Date.now();
+        recordingInProgressRef.current = false;
+      });
+
+      currentService.on('close', () => {
+        if (transcriptionServiceRef.current !== currentService) return;
+        if (!stopRequestedRef.current) {
+          setIsRecording(false);
+          isRecordingRef.current = false;
+          setIsConnecting(false);
+          recordingInProgressRef.current = false;
+          lastFailureTimeRef.current = Date.now();
+        }
+      });
+
+      try {
+        await currentService.connect();
+        if (cleanupIfStopped()) return;
+
+        await recorderRef.current?.start();
+        if (cleanupIfStopped()) return;
+
+        setIsConnecting(false);
+
+        const currentVoiceBackend =
+          settings.experimental.voice?.backend ?? 'gemini-live';
+
+        recorderRef.current?.on('data', (chunk) => {
+          if (currentVoiceBackend === 'gemini-live') {
+            currentService.sendAudioChunk(chunk);
+          }
+        });
+        recorderRef.current?.on('error', (err) => {
+          debugLogger.error('[Voice] Recorder error:', err);
+          stopVoiceRecording();
+          lastFailureTimeRef.current = Date.now();
+        });
+      } catch (err: unknown) {
+        if (transcriptionServiceRef.current !== currentService) return;
+        const message = err instanceof Error ? err.message : String(err);
+        setQueueErrorMessage(`Voice mode failure: ${message}`);
+        setIsRecording(false);
+        isRecordingRef.current = false;
+        setIsConnecting(false);
+        recordingInProgressRef.current = false;
+        lastFailureTimeRef.current = Date.now();
+
+        if (recorderRef.current) {
+          recorderRef.current.stop();
+          recorderRef.current = null;
+        }
+        if (transcriptionServiceRef.current) {
+          transcriptionServiceRef.current.disconnect();
+          transcriptionServiceRef.current = null;
+        }
+      }
+    };
+
+    void startAsync();
+  }, [
+    config,
+    settings.experimental.voice,
+    setQueueErrorMessage,
+    stopVoiceRecording,
+  ]);
+
+  useEffect(
+    () => () => {
+      if (voiceTimeoutRef.current) clearTimeout(voiceTimeoutRef.current);
+      if (recorderRef.current) {
+        recorderRef.current.stop();
+        recorderRef.current = null;
+      }
+      if (transcriptionServiceRef.current) {
+        transcriptionServiceRef.current.disconnect();
+        transcriptionServiceRef.current = null;
+      }
+      if (pttTimerRef.current) clearTimeout(pttTimerRef.current);
+      if (disconnectTimerRef.current) clearTimeout(disconnectTimerRef.current);
+    },
+    [],
+  );
+
+  const handleVoiceInput = useCallback(
+    (key: Key): boolean => {
+      const activeRecording = isRecording || isRecordingRef.current;
+
+      if (activeRecording) {
+        const activationMode =
+          settings.experimental.voice?.activationMode ?? 'push-to-talk';
+
+        if (keyMatchers[Command.ESCAPE](key)) {
+          stopVoiceRecording();
+          return true;
+        }
+
+        if (keyMatchers[Command.VOICE_MODE_PTT](key)) {
+          if (activationMode === 'push-to-talk') {
+            if (pttTimerRef.current) {
+              clearTimeout(pttTimerRef.current);
+            }
+            pttTimerRef.current = setTimeout(() => {
+              stopVoiceRecording();
+              pttTimerRef.current = null;
+            }, RELEASE_DELAY_MS);
+            return true;
+          } else {
+            stopVoiceRecording();
+            return true;
+          }
+        }
+        return true;
+      }
+
+      if (isVoiceModeEnabled) {
+        const activationMode =
+          settings.experimental.voice?.activationMode ?? 'push-to-talk';
+
+        if (keyMatchers[Command.ESCAPE](key) && buffer.text === '') {
+          setVoiceModeEnabled(false);
+          return true;
+        }
+
+        if (keyMatchers[Command.VOICE_MODE_PTT](key)) {
+          if (
+            key.name === 'space' &&
+            !key.ctrl &&
+            !key.alt &&
+            !key.shift &&
+            !key.cmd
+          ) {
+            if (activationMode === 'toggle') {
+              startVoiceRecording();
+              return true;
+            } else {
+              if (pttStateRef.current === 'idle') {
+                buffer.insert(' ');
+                pttStateRef.current = 'possible-hold';
+
+                if (pttTimerRef.current) clearTimeout(pttTimerRef.current);
+                pttTimerRef.current = setTimeout(() => {
+                  pttStateRef.current = 'idle';
+                  pttTimerRef.current = null;
+                }, HOLD_DELAY_MS);
+                return true;
+              } else if (pttStateRef.current === 'possible-hold') {
+                if (pttTimerRef.current) clearTimeout(pttTimerRef.current);
+                buffer.backspace();
+                pttStateRef.current = 'recording';
+                startVoiceRecording();
+
+                pttTimerRef.current = setTimeout(() => {
+                  stopVoiceRecording();
+                  pttTimerRef.current = null;
+                }, RELEASE_DELAY_MS);
+                return true;
+              }
+            }
+          }
+        }
+
+        if (pttStateRef.current === 'possible-hold') {
+          pttStateRef.current = 'idle';
+          if (pttTimerRef.current) {
+            clearTimeout(pttTimerRef.current);
+            pttTimerRef.current = null;
+          }
+        }
+      }
+
+      return false;
+    },
+    [
+      isRecording,
+      isVoiceModeEnabled,
+      settings.experimental.voice,
+      keyMatchers,
+      stopVoiceRecording,
+      startVoiceRecording,
+      buffer,
+      setVoiceModeEnabled,
+    ],
+  );
+
+  return {
+    isRecording,
+    isConnecting,
+    startVoiceRecording,
+    stopVoiceRecording,
+    handleVoiceInput,
+    resetTurnBaseline: () => {
+      turnBaselineRef.current = null;
+    },
+  };
+}

packages/cli/src/ui/hooks/useVoiceModelCommand.ts

@@ -0,0 +1,31 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import { useState, useCallback } from 'react';
+
+interface UseVoiceModelCommandReturn {
+  isVoiceModelDialogOpen: boolean;
+  openVoiceModelDialog: () => void;
+  closeVoiceModelDialog: () => void;
+}
+
+export const useVoiceModelCommand = (): UseVoiceModelCommandReturn => {
+  const [isVoiceModelDialogOpen, setIsVoiceModelDialogOpen] = useState(false);
+
+  const openVoiceModelDialog = useCallback(() => {
+    setIsVoiceModelDialogOpen(true);
+  }, []);
+
+  const closeVoiceModelDialog = useCallback(() => {
+    setIsVoiceModelDialogOpen(false);
+  }, []);
+
+  return {
+    isVoiceModelDialogOpen,
+    openVoiceModelDialog,
+    closeVoiceModelDialog,
+  };
+};

packages/cli/src/ui/key/keyBindings.ts

@@ -97,6 +97,7 @@ export enum Command {
   RESTART_APP = 'app.restart',
   SUSPEND_APP = 'app.suspend',
   SHOW_SHELL_INPUT_UNFOCUS_WARNING = 'app.showShellUnfocusWarning',
+  VOICE_MODE_PTT = 'app.voiceModePTT',
 
   // Background Shell Controls
   BACKGROUND_SHELL_ESCAPE = 'background.escape',
@@ -407,9 +408,7 @@ export const defaultKeyBindingConfig: KeyBindingConfig = new Map([
   [Command.RESTART_APP, [new KeyBinding('r'), new KeyBinding('shift+r')]],
   [Command.SUSPEND_APP, [new KeyBinding('ctrl+z')]],
   [Command.SHOW_SHELL_INPUT_UNFOCUS_WARNING, [new KeyBinding('tab')]],
-  [Command.DUMP_FRAME, [new KeyBinding('f8')]],
-  [Command.START_RECORDING, [new KeyBinding('f6')]],
-  [Command.STOP_RECORDING, [new KeyBinding('f7')]],
+  [Command.VOICE_MODE_PTT, [new KeyBinding('space')]],
 
   // Background Shell Controls
   [Command.BACKGROUND_SHELL_ESCAPE, [new KeyBinding('escape')]],
@@ -424,6 +423,10 @@ export const defaultKeyBindingConfig: KeyBindingConfig = new Map([
   // Extension Controls
   [Command.UPDATE_EXTENSION, [new KeyBinding('i')]],
   [Command.LINK_EXTENSION, [new KeyBinding('l')]],
+
+  [Command.DUMP_FRAME, [new KeyBinding('f8')]],
+  [Command.START_RECORDING, [new KeyBinding('f6')]],
+  [Command.STOP_RECORDING, [new KeyBinding('f7')]],
 ]);
 
 interface CommandCategory {
@@ -538,6 +541,7 @@ export const commandCategories: readonly CommandCategory[] = [
       Command.RESTART_APP,
       Command.SUSPEND_APP,
       Command.SHOW_SHELL_INPUT_UNFOCUS_WARNING,
+      Command.VOICE_MODE_PTT,
     ],
   },
   {
@@ -658,6 +662,7 @@ export const commandDescriptions: Readonly<Record<Command, string>> = {
   [Command.SUSPEND_APP]: 'Suspend the CLI and move it to the background.',
   [Command.SHOW_SHELL_INPUT_UNFOCUS_WARNING]:
     'Show warning when trying to move focus away from shell input.',
+  [Command.VOICE_MODE_PTT]: 'Hold to speak in Voice Mode.',
 
   // Background Shell Controls
   [Command.BACKGROUND_SHELL_ESCAPE]: 'Dismiss background shell list.',

packages/cli/src/ui/noninteractive/nonInteractiveUi.ts

@@ -43,5 +43,6 @@ export function createNonInteractiveUI(): CommandContext['ui'] {
     removeComponent: () => {},
     toggleBackgroundTasks: () => {},
     toggleShortcutsHelp: () => {},
+    toggleVoiceMode: () => {},
   };
 }

packages/cli/src/ui/utils/updateCheck.test.ts

@@ -15,6 +15,25 @@ const debugLogger = vi.hoisted(() => ({
 vi.mock('@google/gemini-cli-core', () => ({
   getPackageJson,
   debugLogger,
+  ReleaseChannel: {
+    NIGHTLY: 'nightly',
+    PREVIEW: 'preview',
+    STABLE: 'stable',
+  },
+  getChannelFromVersion: (version: string) => {
+    if (!version || version.includes('nightly')) {
+      return 'nightly';
+    }
+    if (version.includes('preview')) {
+      return 'preview';
+    }
+    return 'stable';
+  },
+  RELEASE_CHANNEL_STABILITY: {
+    nightly: 0,
+    preview: 1,
+    stable: 2,
+  },
 }));
 
 const latestVersion = vi.hoisted(() => vi.fn());
@@ -152,4 +171,68 @@ describe('checkForUpdates', () => {
       expect(result?.update.latest).toBe('1.2.3-nightly.2');
     });
   });
+
+  describe('channel stability', () => {
+    it('should NOT offer nightly update to a stable user even if tagged as latest', async () => {
+      getPackageJson.mockResolvedValue({
+        name: 'test-package',
+        version: '1.0.0',
+      });
+      // latest points to a nightly that is semver-greater
+      latestVersion.mockResolvedValue('1.1.0-nightly.1');
+
+      const result = await checkForUpdates(mockSettings);
+      expect(result).toBeNull();
+    });
+
+    it('should NOT offer preview update to a stable user even if tagged as latest', async () => {
+      getPackageJson.mockResolvedValue({
+        name: 'test-package',
+        version: '1.0.0',
+      });
+      // latest points to a preview that is semver-greater
+      latestVersion.mockResolvedValue('1.1.0-preview.1');
+
+      const result = await checkForUpdates(mockSettings);
+      expect(result).toBeNull();
+    });
+
+    it('should offer stable update to a stable user', async () => {
+      getPackageJson.mockResolvedValue({
+        name: 'test-package',
+        version: '1.0.0',
+      });
+      latestVersion.mockResolvedValue('1.1.0');
+
+      const result = await checkForUpdates(mockSettings);
+      expect(result?.update.latest).toBe('1.1.0');
+    });
+
+    it('should offer stable update to a nightly user', async () => {
+      getPackageJson.mockResolvedValue({
+        name: 'test-package',
+        version: '1.0.0-nightly.1',
+      });
+      latestVersion.mockImplementation(async (name, options) => {
+        if (options?.version === 'nightly') {
+          return '1.0.0-nightly.1'; // No nightly update
+        }
+        return '1.1.0'; // Stable update available
+      });
+
+      const result = await checkForUpdates(mockSettings);
+      expect(result?.update.latest).toBe('1.1.0');
+    });
+
+    it('should offer stable update to a preview user', async () => {
+      getPackageJson.mockResolvedValue({
+        name: 'test-package',
+        version: '1.0.0-preview.1',
+      });
+      latestVersion.mockResolvedValue('1.1.0');
+
+      const result = await checkForUpdates(mockSettings);
+      expect(result?.update.latest).toBe('1.1.0');
+    });
+  });
 });

packages/cli/src/ui/utils/updateCheck.ts

@@ -6,7 +6,12 @@
 
 import latestVersion from 'latest-version';
 import semver from 'semver';
-import { getPackageJson, debugLogger } from '@google/gemini-cli-core';
+import {
+  getPackageJson,
+  debugLogger,
+  getChannelFromVersion,
+  RELEASE_CHANNEL_STABILITY,
+} from '@google/gemini-cli-core';
 import type { LoadedSettings } from '../../config/settings.js';
 import { fileURLToPath } from 'node:url';
 import path from 'node:path';
@@ -65,6 +70,7 @@ export async function checkForUpdates(
     }
 
     const { name, version: currentVersion } = packageJson;
+    const currentChannel = getChannelFromVersion(currentVersion);
     const isNightly = currentVersion.includes('nightly');
 
     if (isNightly) {
@@ -90,8 +96,21 @@ export async function checkForUpdates(
       }
     } else {
       const latestUpdate = await latestVersion(name);
+      if (!latestUpdate) {
+        return null;
+      }
 
-      if (latestUpdate && semver.gt(latestUpdate, currentVersion)) {
+      const targetChannel = getChannelFromVersion(latestUpdate);
+
+      // Only offer updates that are as stable or more stable than the current version
+      if (
+        RELEASE_CHANNEL_STABILITY[targetChannel] <
+        RELEASE_CHANNEL_STABILITY[currentChannel]
+      ) {
+        return null;
+      }
+
+      if (semver.gt(latestUpdate, currentVersion)) {
         const message = `Gemini CLI update available! ${currentVersion} → ${latestUpdate}`;
         const type = semver.diff(latestUpdate, currentVersion) || undefined;
         return {

packages/cli/src/utils/activityLogger.test.ts

@@ -4,7 +4,7 @@
  * SPDX-License-Identifier: Apache-2.0
  */
 
-import { describe, it, expect, beforeEach } from 'vitest';
+import { describe, it, expect, beforeEach, vi } from 'vitest';
 import { ActivityLogger, type NetworkLog } from './activityLogger.js';
 import type { ConsoleLogPayload } from '@google/gemini-cli-core';
 
@@ -132,4 +132,95 @@ describe('ActivityLogger', () => {
     expect(after.console.length).toBe(0);
     expect(after.network.length).toBe(0);
   });
+
+  it('preserves headers and method from Request object when intercepting fetch', async () => {
+    const originalFetch = global.fetch;
+
+    const mockFetch = vi.fn().mockImplementation(() =>
+      Promise.resolve({
+        status: 200,
+        headers: new Headers(),
+        body: null,
+        clone: () => ({
+          body: null,
+          status: 200,
+          headers: new Headers(),
+          text: async () => 'ok',
+          json: async () => ({}),
+        }),
+      } as unknown as Response),
+    );
+
+    global.fetch = mockFetch;
+
+    try {
+      // @ts-expect-error - accessing private property for testing
+      logger.isInterceptionEnabled = false;
+      logger.enable();
+
+      const request = new Request('https://api.example.com/data', {
+        headers: { Authorization: 'Bearer test-token' },
+        method: 'POST',
+      });
+
+      await global.fetch(request);
+
+      expect(mockFetch).toHaveBeenCalled();
+      const [, calledInit] = mockFetch.mock.calls[0];
+
+      expect(calledInit?.headers).toBeDefined();
+      const headers = new Headers(calledInit?.headers as HeadersInit);
+      expect(headers.get('Authorization')).toBe('Bearer test-token');
+      expect(headers.has('x-activity-request-id')).toBe(true);
+      expect(calledInit?.method).toBe('POST');
+    } finally {
+      global.fetch = originalFetch;
+      // @ts-expect-error - reset private property
+      logger.isInterceptionEnabled = false;
+    }
+  });
+
+  it('replaces Request headers with init headers (Fetch spec compliance)', async () => {
+    const originalFetch = global.fetch;
+    const mockFetch = vi.fn().mockImplementation(() =>
+      Promise.resolve({
+        status: 200,
+        headers: new Headers(),
+        body: null,
+        clone: () => ({
+          body: null,
+          status: 200,
+          headers: new Headers(),
+          text: async () => 'ok',
+        }),
+      } as unknown as Response),
+    );
+    global.fetch = mockFetch;
+
+    try {
+      // @ts-expect-error - accessing private property for testing
+      logger.isInterceptionEnabled = false;
+      logger.enable();
+
+      const request = new Request('https://api.example.com/data', {
+        headers: { 'X-Old': 'old-value', 'X-Shared': 'old-shared' },
+      });
+
+      await global.fetch(request, {
+        headers: { 'X-New': 'new-value', 'X-Shared': 'new-shared' },
+      });
+
+      const [, calledInit] = mockFetch.mock.calls[0];
+      const headers = new Headers(calledInit?.headers as HeadersInit);
+
+      expect(headers.get('X-New')).toBe('new-value');
+      expect(headers.get('X-Shared')).toBe('new-shared');
+      expect(headers.has('X-Old')).toBe(false);
+      expect(headers.has('x-activity-request-id')).toBe(true);
+    } finally {
+      global.fetch = originalFetch;
+      // @ts-expect-error - reset private property
+      logger.isInterceptionEnabled = false;
+    }
+  });
 });

packages/cli/src/utils/activityLogger.ts

@@ -302,18 +302,31 @@ export class ActivityLogger extends EventEmitter {
         return originalFetch(input, init);
 
       const id = Math.random().toString(36).substring(7);
-      const method = (init?.method || 'GET').toUpperCase();
 
-      const newInit = { ...init };
-      const headers = new Headers(init?.headers || {});
+      const inputMethod =
+        typeof input === 'object' && 'method' in input
+          ? input.method
+          : undefined;
+      const inputHeaders =
+        typeof input === 'object' && 'headers' in input
+          ? input.headers
+          : undefined;
+
+      const method = (init?.method ?? inputMethod ?? 'GET').toUpperCase();
+      const headers = new Headers(init?.headers ?? inputHeaders ?? {});
       headers.set(ACTIVITY_ID_HEADER, id);
-      newInit.headers = headers;
+
+      const newInit = {
+        ...init,
+        method,
+        headers,
+      };
 
       let reqBody = '';
-      if (init?.body) {
-        if (typeof init.body === 'string') reqBody = init.body;
-        else if (init.body instanceof URLSearchParams)
-          reqBody = init.body.toString();
+      const body = newInit.body;
+      if (body) {
+        if (typeof body === 'string') reqBody = body;
+        else if (body instanceof URLSearchParams) reqBody = body.toString();
       }
 
       this.requestStartTimes.set(id, Date.now());

packages/cli/src/utils/handleAutoUpdate.test.ts

@@ -301,7 +301,7 @@ describe('handleAutoUpdate', () => {
 
     expect(updateEventEmitter.emit).toHaveBeenCalledWith('update-failed', {
       message:
-        'Automatic update failed. Please try updating manually. (command: npm i -g @google/gemini-cli@2.0.0)',
+        'Automatic update failed. Please try updating manually:\n\nnpm i -g @google/gemini-cli@2.0.0',
     });
   });
 
@@ -325,7 +325,7 @@ describe('handleAutoUpdate', () => {
 
     expect(updateEventEmitter.emit).toHaveBeenCalledWith('update-failed', {
       message:
-        'Automatic update failed. Please try updating manually. (error: Spawn error)',
+        'Automatic update failed. Please try updating manually. (error: Spawn error)\n\nnpm i -g @google/gemini-cli@2.0.0',
     });
   });
 
@@ -334,7 +334,8 @@ describe('handleAutoUpdate', () => {
       ...mockUpdateInfo,
       update: {
         ...mockUpdateInfo.update,
-        latest: '2.0.0-nightly',
+        current: '1.0.0-nightly.0',
+        latest: '2.0.0-nightly.1',
       },
     };
     mockGetInstallationInfo.mockReturnValue({
@@ -356,6 +357,26 @@ describe('handleAutoUpdate', () => {
     );
   });
 
+  it('should NOT update if target is less stable than current (defense-in-depth)', async () => {
+    mockUpdateInfo = {
+      ...mockUpdateInfo,
+      update: {
+        ...mockUpdateInfo.update,
+        current: '1.0.0',
+        latest: '1.1.0-nightly.1',
+      },
+    };
+    mockGetInstallationInfo.mockReturnValue({
+      updateCommand: 'npm i -g @google/gemini-cli@latest',
+      isGlobal: false,
+      packageManager: PackageManager.NPM,
+    });
+
+    handleAutoUpdate(mockUpdateInfo, mockSettings, '/root', mockSpawn);
+
+    expect(mockSpawn).not.toHaveBeenCalled();
+  });
+
   it('should emit "update-success" when the update process succeeds', async () => {
     await new Promise<void>((resolve) => {
       mockGetInstallationInfo.mockReturnValue({
@@ -435,13 +456,15 @@ describe('setUpdateHandler', () => {
   });
 
   it('should handle update-failed event', () => {
-    updateEventEmitter.emit('update-failed', { message: 'Failed' });
+    updateEventEmitter.emit('update-failed', {
+      message: 'Failed message with command',
+    });
 
     expect(setUpdateInfo).toHaveBeenCalledWith(null);
     expect(addItem).toHaveBeenCalledWith(
       {
         type: MessageType.ERROR,
-        text: 'Automatic update failed. Please try updating manually',
+        text: 'Failed message with command',
       },
       expect.any(Number),
     );

packages/cli/src/utils/handleAutoUpdate.ts

@@ -11,7 +11,11 @@ import { updateEventEmitter } from './updateEventEmitter.js';
 import { MessageType, type HistoryItem } from '../ui/types.js';
 import { spawnWrapper } from './spawnWrapper.js';
 import type { spawn } from 'node:child_process';
-import { debugLogger } from '@google/gemini-cli-core';
+import {
+  debugLogger,
+  getChannelFromVersion,
+  RELEASE_CHANNEL_STABILITY,
+} from '@google/gemini-cli-core';
 
 let _updateInProgress = false;
 
@@ -122,6 +126,25 @@ export function handleAutoUpdate(
     return;
   }
 
+  const currentVersion = info.update.current;
+  if (!currentVersion) {
+    debugLogger.warn(
+      'Update check: current version is missing. Skipping automatic update for safety.',
+    );
+    return;
+  }
+
+  const currentChannel = getChannelFromVersion(currentVersion);
+  const targetChannel = getChannelFromVersion(info.update.latest);
+
+  // Defense-in-depth: prevent updates to a less stable channel
+  if (
+    RELEASE_CHANNEL_STABILITY[targetChannel] <
+    RELEASE_CHANNEL_STABILITY[currentChannel]
+  ) {
+    return;
+  }
+
   const isNightly = info.update.latest.includes('nightly');
 
   const updateCommand = installationInfo.updateCommand.replace(
@@ -148,7 +171,7 @@ export function handleAutoUpdate(
       });
     } else {
       updateEventEmitter.emit('update-failed', {
-        message: `Automatic update failed. Please try updating manually. (command: ${updateCommand})`,
+        message: `Automatic update failed. Please try updating manually:\n\n${updateCommand}`,
       });
     }
   });
@@ -156,7 +179,7 @@ export function handleAutoUpdate(
   updateProcess.on('error', (err) => {
     _updateInProgress = false;
     updateEventEmitter.emit('update-failed', {
-      message: `Automatic update failed. Please try updating manually. (error: ${err.message})`,
+      message: `Automatic update failed. Please try updating manually. (error: ${err.message})\n\n${updateCommand}`,
     });
   });
   return updateProcess;
@@ -184,12 +207,14 @@ export function setUpdateHandler(
     }, 60000);
   };
 
-  const handleUpdateFailed = () => {
+  const handleUpdateFailed = (data?: { message: string }) => {
     setUpdateInfo(null);
     addItem(
       {
         type: MessageType.ERROR,
-        text: `Automatic update failed. Please try updating manually`,
+        text:
+          data?.message ||
+          `Automatic update failed. Please try updating manually`,
       },
       Date.now(),
     );

packages/cli/src/utils/processUtils.test.ts

@@ -9,6 +9,11 @@ import {
   RELAUNCH_EXIT_CODE,
   relaunchApp,
   _resetRelaunchStateForTesting,
+  isStandardSea,
+  getScriptArgs,
+  isSeaEnvironment,
+  getSpawnConfig,
+  type ProcessWithSea,
 } from './processUtils.js';
 import * as cleanup from './cleanup.js';
 import * as handleAutoUpdate from './handleAutoUpdate.js';
@@ -36,3 +41,156 @@ describe('processUtils', () => {
     expect(processExit).toHaveBeenCalledWith(RELAUNCH_EXIT_CODE);
   });
 });
+
+describe('SEA handling utilities', () => {
+  const originalArgv = process.argv;
+  const originalExecArgv = process.execArgv;
+  const originalExecPath = process.execPath;
+  const originalIsSea = (process as ProcessWithSea).isSea;
+
+  beforeEach(() => {
+    vi.unstubAllEnvs();
+    vi.stubEnv('NODE_OPTIONS', '');
+    process.argv = [...originalArgv];
+    process.execArgv = [...originalExecArgv];
+    process.execPath = '/fake/exec/path';
+    delete (process as ProcessWithSea).isSea;
+  });
+
+  afterEach(() => {
+    vi.unstubAllEnvs();
+    process.argv = originalArgv;
+    process.execArgv = originalExecArgv;
+    process.execPath = originalExecPath;
+    if (originalIsSea) {
+      (process as ProcessWithSea).isSea = originalIsSea;
+    } else {
+      delete (process as ProcessWithSea).isSea;
+    }
+  });
+
+  describe('isStandardSea', () => {
+    it('returns false if argv[0] === argv[1]', () => {
+      process.argv = ['/bin/gemini', '/bin/gemini', 'my-command'];
+      vi.stubEnv('IS_BINARY', 'true');
+      expect(isStandardSea()).toBe(false);
+    });
+
+    it('returns true if IS_BINARY is true and argv[0] !== argv[1]', () => {
+      process.argv = ['/bin/gemini', 'my-command'];
+      vi.stubEnv('IS_BINARY', 'true');
+      expect(isStandardSea()).toBe(true);
+    });
+
+    it('returns true if process.isSea() is true and argv[0] !== argv[1]', () => {
+      process.argv = ['/bin/gemini', 'my-command'];
+      (process as ProcessWithSea).isSea = () => true;
+      expect(isStandardSea()).toBe(true);
+    });
+
+    it('returns false in standard node environment', () => {
+      process.argv = ['/bin/node', '/path/to/script.js', 'my-command'];
+      expect(isStandardSea()).toBe(false);
+    });
+  });
+
+  describe('getScriptArgs', () => {
+    it('slices from index 1 if isStandardSea is true', () => {
+      process.argv = ['/bin/gemini', 'my-command', '--flag'];
+      vi.stubEnv('IS_BINARY', 'true');
+      expect(getScriptArgs()).toEqual(['my-command', '--flag']);
+    });
+
+    it('slices from index 2 if isStandardSea is false (relaunch SEA or standard node)', () => {
+      // Relaunch SEA
+      process.argv = ['/bin/gemini', '/bin/gemini', 'my-command', '--flag'];
+      vi.stubEnv('IS_BINARY', 'true');
+      expect(getScriptArgs()).toEqual(['my-command', '--flag']);
+
+      // Standard node
+      process.argv = ['/bin/node', '/path/to/script.js', 'my-command'];
+      vi.stubEnv('IS_BINARY', '');
+      expect(getScriptArgs()).toEqual(['my-command']);
+    });
+  });
+
+  describe('isSeaEnvironment', () => {
+    it('returns true if IS_BINARY is true', () => {
+      vi.stubEnv('IS_BINARY', 'true');
+      expect(isSeaEnvironment()).toBe(true);
+    });
+
+    it('returns true if process.isSea() is true', () => {
+      (process as ProcessWithSea).isSea = () => true;
+      expect(isSeaEnvironment()).toBe(true);
+    });
+
+    it('returns true if argv[0] === argv[1]', () => {
+      process.argv = ['/bin/gemini', '/bin/gemini'];
+      expect(isSeaEnvironment()).toBe(true);
+    });
+
+    it('returns false otherwise', () => {
+      process.argv = ['/bin/node', '/path/to/script.js'];
+      expect(isSeaEnvironment()).toBe(false);
+    });
+  });
+
+  describe('getSpawnConfig', () => {
+    it('handles standard node mode', () => {
+      process.argv = ['/bin/node', '/path/to/script.js', 'my-command'];
+      process.execArgv = ['--inspect'];
+      process.execPath = '/bin/node';
+
+      const config = getSpawnConfig(
+        ['--max-old-space-size=8192'],
+        ['my-command'],
+      );
+
+      expect(config.spawnArgs).toEqual([
+        '--inspect',
+        '--max-old-space-size=8192',
+        '/path/to/script.js',
+        'my-command',
+      ]);
+      expect(config.env['GEMINI_CLI_NO_RELAUNCH']).toBe('true');
+      expect(config.env['NODE_OPTIONS']).toBeFalsy();
+    });
+
+    it('handles SEA binary mode with new nodeArgs', () => {
+      vi.stubEnv('IS_BINARY', 'true');
+      vi.stubEnv('NODE_OPTIONS', '--existing-flag');
+      process.argv = ['/bin/gemini', 'my-command'];
+      process.execArgv = ['--inspect']; // Should not be duplicated in NODE_OPTIONS
+      process.execPath = '/bin/gemini';
+
+      const config = getSpawnConfig(
+        ['--max-old-space-size=8192'],
+        ['my-command'],
+      );
+
+      expect(config.spawnArgs).toEqual([
+        '/bin/gemini', // explicitly uses execPath as placeholder
+        'my-command',
+      ]);
+      expect(config.env['NODE_OPTIONS']).toBe(
+        '--existing-flag --max-old-space-size=8192',
+      );
+      expect(config.env['GEMINI_CLI_NO_RELAUNCH']).toBe('true');
+    });
+
+    it('throws error for complex nodeArgs in SEA mode', () => {
+      vi.stubEnv('IS_BINARY', 'true');
+
+      expect(() => {
+        getSpawnConfig(['--title "My App"'], []);
+      }).toThrow(
+        'Unsupported node argument for SEA relaunch: --title "My App". Complex escaping is not supported.',
+      );
+
+      expect(() => {
+        getSpawnConfig(['--title=A\\B'], []);
+      }).toThrow();
+    });
+  });
+});

packages/cli/src/utils/processUtils.ts

@@ -29,3 +29,101 @@ export async function relaunchApp(): Promise<void> {
   await runExitCleanup();
   process.exit(RELAUNCH_EXIT_CODE);
 }
+
+export interface ProcessWithSea extends NodeJS.Process {
+  isSea?: () => boolean;
+}
+
+/**
+ * Determines whether the current process is a "standard" SEA (Single Executable Application)
+ * where the user arguments start at index 1 instead of index 2.
+ * A relaunched SEA child will have process.argv[0] === process.argv[1] (because we inject execPath),
+ * so it will return false here and correctly slice from index 2.
+ */
+export function isStandardSea(): boolean {
+  return (
+    process.argv[0] !== process.argv[1] &&
+    (process.env['IS_BINARY'] === 'true' ||
+      (process as ProcessWithSea).isSea?.() === true)
+  );
+}
+
+/**
+ * Extracts the user-provided script arguments from process.argv,
+ * accounting for the differences in SEA execution modes.
+ */
+export function getScriptArgs(): string[] {
+  return process.argv.slice(isStandardSea() ? 1 : 2);
+}
+
+/**
+ * Determines if the current process is running in any SEA environment
+ * (either the initial launch or a relaunched child).
+ */
+export function isSeaEnvironment(): boolean {
+  return (
+    process.env['IS_BINARY'] === 'true' ||
+    (process as ProcessWithSea).isSea?.() === true ||
+    process.argv[0] === process.argv[1]
+  );
+}
+
+/**
+ * Constructs the arguments and environment for spawning a child process during relaunch.
+ * Handles differences between standard Node and SEA binary modes.
+ */
+export function getSpawnConfig(
+  nodeArgs: string[],
+  scriptArgs: string[],
+): {
+  spawnArgs: string[];
+  env: NodeJS.ProcessEnv;
+} {
+  const isBinary = isSeaEnvironment();
+  const newEnv: NodeJS.ProcessEnv = {
+    ...process.env,
+    GEMINI_CLI_NO_RELAUNCH: 'true',
+  };
+
+  const finalSpawnArgs: string[] = [];
+
+  if (isBinary) {
+    // In SEA mode, Node flags must be passed via NODE_OPTIONS, as the binary
+    // passes all CLI arguments directly to the application.
+    // We only need to append the *new* nodeArgs (e.g., memory flags).
+    // Existing execArgv are inherited via the environment or baked into the binary.
+    if (nodeArgs.length > 0) {
+      for (const arg of nodeArgs) {
+        if (/[\s"'\\]/.test(arg)) {
+          throw new Error(
+            `Unsupported node argument for SEA relaunch: ${arg}. Complex escaping is not supported.`,
+          );
+        }
+      }
+      const existingNodeOptions = process.env['NODE_OPTIONS'] || '';
+      // nodeArgs in our codebase are simple flags like --max-old-space-size=X
+      // that do not contain spaces and do not require complex escaping.
+      newEnv['NODE_OPTIONS'] =
+        `${existingNodeOptions} ${nodeArgs.join(' ')}`.trim();
+    }
+    // Binary is its own entry point. To maintain the [node, script, ...args]
+    // structure expected by the application (which uses slice(2)),
+    // we must provide a placeholder for the script path.
+    // We explicitly use process.execPath to break the cycle and prevent
+    // compounding argument duplication on subsequent relaunches.
+    finalSpawnArgs.push(process.execPath, ...scriptArgs);
+  } else {
+    // Standard Node mode: pass all flags via command line.
+    finalSpawnArgs.push(
+      ...process.execArgv,
+      ...nodeArgs,
+      process.argv[1],
+      ...scriptArgs,
+    );
+  }
+
+  return {
+    spawnArgs: finalSpawnArgs,
+    env: newEnv,
+  };
+}

packages/cli/src/utils/relaunch.test.ts

@@ -59,6 +59,7 @@ describe('relaunchOnExitCode', () => {
   });
 
   afterEach(() => {
+    vi.unstubAllEnvs();
     processExitSpy.mockRestore();
     stdinResumeSpy.mockRestore();
   });
@@ -116,7 +117,6 @@ describe('relaunchAppInChildProcess', () => {
   let stdinResumeSpy: MockInstance;
 
   // Store original values to restore later
-  const originalEnv = { ...process.env };
   const originalExecArgv = [...process.execArgv];
   const originalArgv = [...process.argv];
   const originalExecPath = process.execPath;
@@ -125,8 +125,9 @@ describe('relaunchAppInChildProcess', () => {
     vi.clearAllMocks();
     mocks.writeToStderr.mockClear();
 
-    process.env = { ...originalEnv };
-    delete process.env['GEMINI_CLI_NO_RELAUNCH'];
+    vi.stubEnv('GEMINI_CLI_NO_RELAUNCH', '');
+    vi.stubEnv('IS_BINARY', '');
+    vi.stubEnv('NODE_OPTIONS', '');
 
     process.execArgv = [...originalExecArgv];
     process.argv = [...originalArgv];
@@ -144,7 +145,7 @@ describe('relaunchAppInChildProcess', () => {
   });
 
   afterEach(() => {
-    process.env = { ...originalEnv };
+    vi.unstubAllEnvs();
     process.execArgv = [...originalExecArgv];
     process.argv = [...originalArgv];
     process.execPath = originalExecPath;
@@ -156,7 +157,7 @@ describe('relaunchAppInChildProcess', () => {
 
   describe('when GEMINI_CLI_NO_RELAUNCH is set', () => {
     it('should return early without spawning a child process', async () => {
-      process.env['GEMINI_CLI_NO_RELAUNCH'] = 'true';
+      vi.stubEnv('GEMINI_CLI_NO_RELAUNCH', 'true');
 
       await relaunchAppInChildProcess(['--test'], ['--verbose']);
 
@@ -167,132 +168,141 @@ describe('relaunchAppInChildProcess', () => {
 
   describe('when GEMINI_CLI_NO_RELAUNCH is not set', () => {
     beforeEach(() => {
-      delete process.env['GEMINI_CLI_NO_RELAUNCH'];
+      vi.stubEnv('GEMINI_CLI_NO_RELAUNCH', '');
     });
 
-    it('should construct correct node arguments from execArgv, additionalNodeArgs, script, additionalScriptArgs, and argv', () => {
-      // Test the argument construction logic directly by extracting it into a testable function
-      // This tests the same logic that's used in relaunchAppInChildProcess
-
-      // Setup test data to verify argument ordering
-      const mockExecArgv = ['--inspect=9229', '--trace-warnings'];
-      const mockArgv = [
+    it('should construct correct spawn arguments and use command line for node arguments in standard Node mode', async () => {
+      process.execArgv = ['--inspect=9229', '--trace-warnings'];
+      process.argv = [
         '/usr/bin/node',
         '/path/to/cli.js',
         'command',
         '--flag=value',
         '--verbose',
       ];
+
       const additionalNodeArgs = [
         '--max-old-space-size=4096',
         '--experimental-modules',
       ];
       const additionalScriptArgs = ['--model', 'gemini-1.5-pro', '--debug'];
 
-      // Extract the argument construction logic from relaunchAppInChildProcess
-      const script = mockArgv[1];
-      const scriptArgs = mockArgv.slice(2);
+      const mockChild = createMockChildProcess(0, true);
+      mockedSpawn.mockReturnValue(mockChild);
 
-      const nodeArgs = [
-        ...mockExecArgv,
-        ...additionalNodeArgs,
-        script,
-        ...additionalScriptArgs,
-        ...scriptArgs,
+      await expect(
+        relaunchAppInChildProcess(additionalNodeArgs, additionalScriptArgs),
+      ).rejects.toThrow('PROCESS_EXIT_CALLED');
+
+      expect(mockedSpawn).toHaveBeenCalledWith(
+        process.execPath,
+        [
+          '--inspect=9229',
+          '--trace-warnings',
+          '--max-old-space-size=4096',
+          '--experimental-modules',
+          '/path/to/cli.js',
+          '--model',
+          'gemini-1.5-pro',
+          '--debug',
+          'command',
+          '--flag=value',
+          '--verbose',
+        ],
+        expect.objectContaining({
+          env: expect.objectContaining({
+            GEMINI_CLI_NO_RELAUNCH: 'true',
+          }),
+        }),
+      );
+
+      const lastCall = mockedSpawn.mock.calls[0] as unknown as [
+        string,
+        string[],
+        { env: NodeJS.ProcessEnv },
       ];
+      const env = lastCall[2].env;
+      expect(env['NODE_OPTIONS']).toBeFalsy();
+    });
 
-      // Verify the argument construction follows the expected pattern:
-      // [...process.execArgv, ...additionalNodeArgs, script, ...additionalScriptArgs, ...scriptArgs]
-      const expectedArgs = [
-        // Original node execution arguments
-        '--inspect=9229',
-        '--trace-warnings',
-        // Additional node arguments passed to function
-        '--max-old-space-size=4096',
-        '--experimental-modules',
-        // The script path
-        '/path/to/cli.js',
-        // Additional script arguments passed to function
-        '--model',
-        'gemini-1.5-pro',
-        '--debug',
-        // Original script arguments (everything after the script in process.argv)
+    it('should handle SEA binary mode (IS_BINARY=true) correctly using NODE_OPTIONS', async () => {
+      vi.stubEnv('IS_BINARY', 'true');
+      // execArgv should be inherited, not duplicated in NODE_OPTIONS
+      process.execArgv = ['--inspect=9229'];
+      process.argv = [
+        '/usr/bin/gemini',
+        '/usr/bin/gemini',
         'command',
-        '--flag=value',
         '--verbose',
       ];
 
-      expect(nodeArgs).toEqual(expectedArgs);
-    });
-
-    it('should handle empty additional arguments correctly', () => {
-      // Test edge cases with empty arrays
-      const mockExecArgv = ['--trace-warnings'];
-      const mockArgv = ['/usr/bin/node', '/app/cli.js', 'start'];
-      const additionalNodeArgs: string[] = [];
+      const additionalNodeArgs = ['--max-old-space-size=8192'];
       const additionalScriptArgs: string[] = [];
 
-      // Extract the argument construction logic
-      const script = mockArgv[1];
-      const scriptArgs = mockArgv.slice(2);
+      const mockChild = createMockChildProcess(0, true);
+      mockedSpawn.mockReturnValue(mockChild);
 
-      const nodeArgs = [
-        ...mockExecArgv,
-        ...additionalNodeArgs,
-        script,
-        ...additionalScriptArgs,
-        ...scriptArgs,
-      ];
+      await expect(
+        relaunchAppInChildProcess(additionalNodeArgs, additionalScriptArgs),
+      ).rejects.toThrow('PROCESS_EXIT_CALLED');
 
-      const expectedArgs = ['--trace-warnings', '/app/cli.js', 'start'];
-
-      expect(nodeArgs).toEqual(expectedArgs);
+      expect(mockedSpawn).toHaveBeenCalledWith(
+        process.execPath,
+        ['/usr/bin/node', 'command', '--verbose'],
+        expect.objectContaining({
+          env: expect.objectContaining({
+            GEMINI_CLI_NO_RELAUNCH: 'true',
+            NODE_OPTIONS: '--max-old-space-size=8192',
+          }),
+        }),
+      );
     });
 
-    it('should handle complex argument patterns', () => {
-      // Test with various argument types including flags with values, boolean flags, etc.
-      const mockExecArgv = ['--max-old-space-size=8192'];
-      const mockArgv = [
-        '/usr/bin/node',
-        '/cli.js',
-        '--config=/path/to/config.json',
-        '--verbose',
-        'subcommand',
-        '--output',
-        'file.txt',
-      ];
-      const additionalNodeArgs = ['--inspect-brk=9230'];
-      const additionalScriptArgs = ['--model=gpt-4', '--temperature=0.7'];
+    it('should append new nodeArgs to NODE_OPTIONS in SEA mode without escaping', async () => {
+      vi.stubEnv('IS_BINARY', 'true');
+      vi.stubEnv('NODE_OPTIONS', '--existing-flag');
+      process.execArgv = ['--inspect']; // inherited from env/binary, should not be duplicated
+      process.argv = ['/usr/bin/gemini', '/usr/bin/gemini', 'command'];
 
-      const script = mockArgv[1];
-      const scriptArgs = mockArgv.slice(2);
+      // In our use case, these are simple flags like --max-old-space-size=X
+      const additionalNodeArgs = ['--max-old-space-size=8192'];
+      const additionalScriptArgs: string[] = [];
 
-      const nodeArgs = [
-        ...mockExecArgv,
-        ...additionalNodeArgs,
-        script,
-        ...additionalScriptArgs,
-        ...scriptArgs,
-      ];
+      const mockChild = createMockChildProcess(0, true);
+      mockedSpawn.mockReturnValue(mockChild);
 
-      const expectedArgs = [
-        '--max-old-space-size=8192',
-        '--inspect-brk=9230',
-        '/cli.js',
-        '--model=gpt-4',
-        '--temperature=0.7',
-        '--config=/path/to/config.json',
-        '--verbose',
-        'subcommand',
-        '--output',
-        'file.txt',
-      ];
+      await expect(
+        relaunchAppInChildProcess(additionalNodeArgs, additionalScriptArgs),
+      ).rejects.toThrow('PROCESS_EXIT_CALLED');
 
-      expect(nodeArgs).toEqual(expectedArgs);
+      expect(mockedSpawn).toHaveBeenCalledWith(
+        process.execPath,
+        ['/usr/bin/node', 'command'],
+        expect.objectContaining({
+          env: expect.objectContaining({
+            NODE_OPTIONS: '--existing-flag --max-old-space-size=8192',
+          }),
+        }),
+      );
     });
 
-    // Note: Additional integration tests for spawn behavior are complex due to module mocking
-    // limitations with ES modules. The core logic is tested in relaunchOnExitCode tests.
+    it('should handle empty additional arguments correctly in Node mode', async () => {
+      process.execArgv = ['--trace-warnings'];
+      process.argv = ['/usr/bin/node', '/app/cli.js', 'start'];
+
+      const mockChild = createMockChildProcess(0, true);
+      mockedSpawn.mockReturnValue(mockChild);
+
+      await expect(relaunchAppInChildProcess([], [])).rejects.toThrow(
+        'PROCESS_EXIT_CALLED',
+      );
+
+      expect(mockedSpawn).toHaveBeenCalledWith(
+        process.execPath,
+        ['--trace-warnings', '/app/cli.js', 'start'],
+        expect.anything(),
+      );
+    });
 
     it('should handle null exit code from child process', async () => {
       process.argv = ['/usr/bin/node', '/app/cli.js'];
@@ -342,6 +352,8 @@ function createMockChildProcess(
     disconnect: vi.fn(),
     unref: vi.fn(),
     ref: vi.fn(),
+    on: mockChild.on.bind(mockChild),
+    emit: mockChild.emit.bind(mockChild),
   });
 
   if (autoClose) {