fix: correctly disconnect extension-backed MCP clients when stopping extension (#24050)

Co-authored-by: gemini-cli-robot <224641728+gemini-cli-robot@users.noreply.github.com>
Co-authored-by: g-samroberts <samroberts@google.com>
Co-authored-by: Sam Roberts <158088236+g-samroberts@users.noreply.github.com>

.gemini/settings.json

@@ -2,8 +2,11 @@
   "experimental": {
     "extensionReloading": true,
     "modelSteering": true,
+    "autoMemory": true,
+    "gemma": true,
     "memoryManager": true,
-    "topicUpdateNarration": true
+    "topicUpdateNarration": true,
+    "voiceMode": true
   },
   "general": {
     "devtools": true

.gemini/skills/docs-writer/SKILL.md

@@ -85,17 +85,25 @@ accessible.
 
 - **Callouts**: Use GitHub-flavored markdown alerts to highlight important
   information. To ensure the formatting is preserved by `npm run format`, place
-  an empty line, then the `<!-- prettier-ignore -->` comment directly before
-  the callout block. The callout type (`[!TYPE]`) should be on the first line,
-  followed by a newline, and then the content, with each subsequent line of
-  content starting with `>`. Available types are `NOTE`, `TIP`, `IMPORTANT`,
-  `WARNING`, and `CAUTION`.
+  an empty line, then a prettier ignore comment directly before the callout
+  block. Use `<!-- prettier-ignore -->` for standard Markdown files (`.md`) and
+  `{/* prettier-ignore */}` for MDX files (`.mdx`). The callout type (`[!TYPE]`)
+  should be on the first line, followed by a newline, and then the content, with
+  each subsequent line of content starting with `>`. Available types are `NOTE`,
+  `TIP`, `IMPORTANT`, `WARNING`, and `CAUTION`.
 
-  Example:
+  Example (.md):
 
 <!-- prettier-ignore -->
 > [!NOTE]
 > This is an example of a multi-line note that will be preserved
+> by Prettier.
+
+  Example (.mdx):
+
+{/* prettier-ignore */}
+> [!NOTE]
+> This is an example of a multi-line note that will be preserved
 > by Prettier.
 
 ### Links
@@ -118,6 +126,7 @@ accessible.
 <!-- prettier-ignore -->
 > [!NOTE]
 > This is an experimental feature currently under active development.
+(Note: Use `{/* prettier-ignore */}` if editing an `.mdx` file.)
 
 - **Headings:** Use hierarchical headings to support the user journey.
 - **Procedures:**

.github/actions/run-tests/action.yml

@@ -28,6 +28,7 @@ runs:
     - name: 'Run Tests'
       env:
         GEMINI_API_KEY: '${{ inputs.gemini_api_key }}'
+        GEMINI_CLI_TRUST_WORKSPACE: true
       working-directory: '${{ inputs.working-directory }}'
       run: |-
         echo "::group::Build"

.github/actions/verify-release/action.yml

@@ -98,6 +98,7 @@ runs:
       working-directory: '${{ inputs.working-directory }}'
       env:
         GEMINI_API_KEY: '${{ inputs.gemini_api_key }}'
+        GEMINI_CLI_TRUST_WORKSPACE: true
         INTEGRATION_TEST_USE_INSTALLED_GEMINI: 'true'
         # We must diable CI mode here because it interferes with interactive tests.
         # See https://github.com/google-gemini/gemini-cli/issues/10517

.github/workflows/chained_e2e.yml

@@ -167,6 +167,7 @@ jobs:
       - name: 'Run E2E tests'
         env:
           GEMINI_API_KEY: '${{ secrets.GEMINI_API_KEY }}'
+          GEMINI_CLI_TRUST_WORKSPACE: true
           KEEP_OUTPUT: 'true'
           VERBOSE: 'true'
           BUILD_SANDBOX_FLAGS: '--cache-from type=gha --cache-to type=gha,mode=max'
@@ -212,6 +213,7 @@ jobs:
         if: "${{runner.os != 'Windows'}}"
         env:
           GEMINI_API_KEY: '${{ secrets.GEMINI_API_KEY }}'
+          GEMINI_CLI_TRUST_WORKSPACE: true
           KEEP_OUTPUT: 'true'
           SANDBOX: 'sandbox:none'
           VERBOSE: 'true'
@@ -288,6 +290,7 @@ jobs:
       - name: 'Run E2E tests'
         env:
           GEMINI_API_KEY: '${{ secrets.GEMINI_API_KEY }}'
+          GEMINI_CLI_TRUST_WORKSPACE: true
           KEEP_OUTPUT: 'true'
           SANDBOX: 'sandbox:none'
           VERBOSE: 'true'
@@ -334,6 +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/ci.yml

@@ -102,6 +102,12 @@ jobs:
       - name: 'Run yamllint'
         run: 'node scripts/lint.js --yamllint'
 
+      - name: 'Build project for typecheck'
+        run: 'npm run build'
+
+      - name: 'Run typecheck'
+        run: 'npm run typecheck'
+
       - name: 'Run Prettier'
         run: 'node scripts/lint.js --prettier'
 
@@ -173,6 +179,7 @@ jobs:
       - name: 'Run tests and generate reports'
         env:
           NO_COLOR: true
+          GEMINI_CLI_TRUST_WORKSPACE: true
         run: |
           if [[ "${{ matrix.shard }}" == "cli" ]]; then
             npm run test:ci --workspace "@google/gemini-cli"
@@ -261,6 +268,7 @@ jobs:
       - name: 'Run tests and generate reports'
         env:
           NO_COLOR: true
+          GEMINI_CLI_TRUST_WORKSPACE: true
         run: |
           if [[ "${{ matrix.shard }}" == "cli" ]]; then
             npm run test:ci --workspace "@google/gemini-cli" -- --coverage.enabled=false
@@ -424,6 +432,7 @@ jobs:
         env:
           GEMINI_API_KEY: '${{ secrets.GEMINI_API_KEY }}'
           NO_COLOR: true
+          GEMINI_CLI_TRUST_WORKSPACE: true
           NODE_OPTIONS: '--max-old-space-size=32768 --max-semi-space-size=256'
           UV_THREADPOOL_SIZE: '32'
           NODE_ENV: 'test'

.github/workflows/deflake.yml

@@ -62,6 +62,7 @@ jobs:
       - name: 'Run E2E tests'
         env:
           GEMINI_API_KEY: '${{ secrets.GEMINI_API_KEY }}'
+          GEMINI_CLI_TRUST_WORKSPACE: true
           IS_DOCKER: "${{ matrix.sandbox == 'sandbox:docker' }}"
           KEEP_OUTPUT: 'true'
           RUNS: '${{ github.event.inputs.runs }}'
@@ -105,6 +106,7 @@ jobs:
         if: "runner.os != 'Windows'"
         env:
           GEMINI_API_KEY: '${{ secrets.GEMINI_API_KEY }}'
+          GEMINI_CLI_TRUST_WORKSPACE: true
           KEEP_OUTPUT: 'true'
           RUNS: '${{ github.event.inputs.runs }}'
           SANDBOX: 'sandbox:none'
@@ -159,6 +161,7 @@ jobs:
       - name: 'Run E2E tests'
         env:
           GEMINI_API_KEY: '${{ secrets.GEMINI_API_KEY }}'
+          GEMINI_CLI_TRUST_WORKSPACE: true
           KEEP_OUTPUT: 'true'
           SANDBOX: 'sandbox:none'
           VERBOSE: 'true'

.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

@@ -0,0 +1,221 @@
+name: '🧠 Gemini CLI Bot: Brain'
+
+on:
+  schedule:
+    - cron: '0 0 * * *' # Every 24 hours
+  workflow_dispatch:
+    inputs:
+      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 }}'
+  cancel-in-progress: true
+
+jobs:
+  reasoning:
+    name: 'Brain (Reasoning Layer)'
+    runs-on: 'ubuntu-latest'
+    if: "github.repository == 'google-gemini/gemini-cli'"
+    # 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
+        with:
+          node-version: '20'
+          cache: 'npm'
+
+      - name: 'Install dependencies'
+        run: 'npm ci'
+
+      - name: 'Build Gemini CLI'
+        run: 'npm run bundle'
+
+      - 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' }}"
+        run: 'node bundle/gemini.js --policy tools/gemini-cli-bot/ci-policy.toml -p "$(cat tools/gemini-cli-bot/brain/metrics.md)"'
+
+      - name: 'Run Critique Phase'
+        if: "${{ github.event.inputs.enable_prs == 'true' }}"
+        env:
+          GEMINI_API_KEY: '${{ secrets.GEMINI_API_KEY }}'
+          # This token is strictly readonly as enforced by the job-level permissions.
+          GITHUB_TOKEN: '${{ 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 "$(cat tools/gemini-cli-bot/brain/critique.md)" 2>&1 | tee critique_output.log
+
+             # PIPESTATUS[0] captures the exit code of the node command before the pipe
+             if [ "${PIPESTATUS[0]}" -ne 0 ] || grep -q "\[REJECTED\]" critique_output.log; then
+               echo "Critique failed or rejected changes. Skipping PR creation."
+               echo "[REJECTED]" > critique_result.txt
+             else
+               echo "[APPROVED]" > critique_result.txt
+             fi
+          fi
+
+      - name: 'Generate Patch'
+        if: "${{ github.event.inputs.enable_prs == 'true' }}"
+        run: |
+          touch bot-changes.patch
+          touch pr-description.md
+          if [ -f critique_result.txt ] && grep -q "\[REJECTED\]" critique_result.txt; then
+             echo "Critique rejected. Skipping patch generation."
+          else
+             git diff --staged > bot-changes.patch
+          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
+          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: 'brain-data'
+          path: '${{ runner.temp }}/brain-data/'
+
+      - name: 'Create or Update PR'
+        if: "${{ github.event.inputs.enable_prs == '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 "${{ runner.temp }}/brain-data/branch-name.txt")
+            fi
+
+            # SECURITY: Only allow pushing to branches starting with 'bot/'
+            if [[ ! "$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 "${{ runner.temp }}/brain-data/pr-description.md"
+            else
+              git commit -m "🤖 Gemini Bot Productivity Optimizations"
+            fi
+
+            # Use force to update existing PR branches
+            git push origin "$BRANCH_NAME" --force
+
+            PR_TITLE="🤖 Gemini Bot Productivity Optimizations"
+            if [ -s "${{ runner.temp }}/brain-data/pr-description.md" ]; then
+              PR_TITLE=$(head -n 1 "${{ runner.temp }}/brain-data/pr-description.md")
+            fi
+
+            # Create PR if it doesn't exist
+            if ! gh pr view "$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 Comment'
+        if: "${{ github.event.inputs.enable_prs == 'true' }}"
+        env:
+          GH_TOKEN: '${{ secrets.GEMINI_CLI_ROBOT_GITHUB_PAT }}'
+        run: |
+          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")
+
+            # SECURITY: Only allow commenting on PRs authored by the bot
+            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

@@ -0,0 +1,48 @@
+name: '🔄 Gemini CLI Bot: Pulse'
+
+on:
+  schedule:
+    - cron: '*/30 * * * *' # Every 30 minutes
+  workflow_dispatch:
+
+concurrency:
+  group: '${{ github.workflow }}-${{ github.ref }}'
+  cancel-in-progress: true
+
+permissions:
+  contents: 'write'
+  issues: 'write'
+  pull-requests: 'write'
+
+jobs:
+  pulse:
+    name: 'Pulse (Reflex Layer)'
+    runs-on: 'ubuntu-latest'
+    if: "github.repository == 'google-gemini/gemini-cli'"
+    steps:
+      - name: 'Checkout'
+        uses: 'actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8' # ratchet:actions/checkout@v5
+        with:
+          fetch-depth: 0
+
+      - name: 'Setup Node.js'
+        uses: 'actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020' # ratchet:actions/setup-node@v4
+        with:
+          node-version: '20'
+          cache: 'npm'
+
+      - name: 'Install dependencies'
+        run: 'npm ci'
+
+      - name: 'Run Reflex Processes'
+        env:
+          GITHUB_TOKEN: '${{ secrets.GITHUB_TOKEN }}'
+        run: |
+          if [ -d "tools/gemini-cli-bot/reflexes/scripts" ] && [ "$(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
+          else
+            echo "No reflex scripts found."
+          fi

.github/workflows/test-build-binary.yml

@@ -141,6 +141,7 @@ jobs:
         if: "github.event_name != 'pull_request'"
         env:
           GEMINI_API_KEY: '${{ secrets.GEMINI_API_KEY }}'
+          GEMINI_CLI_TRUST_WORKSPACE: true
         run: |
           echo "Running integration tests with binary..."
           if [[ "${{ matrix.os }}" == 'windows-latest' ]]; then

.prettierignore

@@ -22,3 +22,4 @@ Thumbs.db
 .pytest_cache
 **/SKILL.md
 packages/sdk/test-data/*.json
+*.mdx

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'));" \

README.md

@@ -371,6 +371,8 @@ for planned features and priorities.
 
 ## 📖 Resources
 
+- **[Free Course](https://learn.deeplearning.ai/courses/gemini-cli-code-and-create-with-an-open-source-agent/information)** -
+  Learn the basics.
 - **[Official Roadmap](./ROADMAP.md)** - See what's coming next.
 - **[Changelog](https://www.geminicli.com/docs/changelogs)** - See recent
   notable updates.

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: v0.38.1
+# Latest stable release: v0.39.0
 
-Released: April 15, 2026
+Released: April 23, 2026
 
 For most users, our latest stable release is the recommended release. Install
 the latest stable version with:
@@ -11,261 +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 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...v0.38.1
+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/auto-memory.md

@@ -0,0 +1,143 @@
+# Auto Memory
+
+Auto Memory is an experimental feature that mines your past Gemini CLI sessions
+in the background and turns recurring workflows into reusable
+[Agent Skills](./skills.md). You review, accept, or discard each extracted skill
+before it becomes available to future sessions.
+
+<!-- prettier-ignore -->
+> [!NOTE]
+> This is an experimental feature currently under active development.
+
+## Overview
+
+Every session you run with Gemini CLI is recorded locally as a transcript. Auto
+Memory scans those transcripts for procedural patterns that recur across
+sessions, then drafts each pattern as a `SKILL.md` file in a project-local
+inbox. You inspect the draft, decide whether it captures real expertise, and
+promote it to your global or workspace skills directory if you want it.
+
+You'll use Auto Memory when you want to:
+
+- **Capture team workflows** that you find yourself walking the agent through
+  more than once.
+- **Codify hard-won fixes** for project-specific landmines so future sessions
+  avoid them.
+- **Bootstrap a skills library** without writing every `SKILL.md` by hand.
+
+Auto Memory complements—but does not replace—the
+[`save_memory` tool](../tools/memory.md), which captures single facts into
+`GEMINI.md`. Auto Memory captures multi-step procedures into skills.
+
+## Prerequisites
+
+- Gemini CLI installed and authenticated.
+- At least 10 user messages across recent, idle sessions in the project. Auto
+  Memory ignores active or trivial sessions.
+
+## How to enable Auto Memory
+
+Auto Memory is off by default. Enable it in your settings file:
+
+1.  Open your global settings file at `~/.gemini/settings.json`. If you only
+    want Auto Memory in one project, edit `.gemini/settings.json` in that
+    project instead.
+
+2.  Add the experimental flag:
+
+    ```json
+    {
+      "experimental": {
+        "autoMemory": true
+      }
+    }
+    ```
+
+3.  Restart Gemini CLI. The flag requires a restart because the extraction
+    service starts during session boot.
+
+## How Auto Memory works
+
+Auto Memory runs as a background task on session startup. It does not block the
+UI, consume your interactive turns, or surface tool prompts.
+
+1.  **Eligibility scan.** The service indexes recent sessions from
+    `~/.gemini/tmp/<project>/chats/`. Sessions are eligible only if they have
+    been idle for at least three hours and contain at least 10 user messages.
+2.  **Lock acquisition.** A lock file in the project's memory directory
+    coordinates across multiple CLI instances so extraction runs at most once at
+    a time.
+3.  **Sub-agent extraction.** A specialized sub-agent (named `confucius`)
+    reviews the session index, reads any sessions that look like they contain
+    repeated procedural workflows, and drafts new `SKILL.md` files. Its
+    instructions tell it to default to creating zero skills unless the evidence
+    is strong, so most runs produce no inbox items.
+4.  **Patch validation.** If the sub-agent proposes edits to skills outside the
+    inbox (for example, an existing global skill), it writes a unified diff
+    `.patch` file. Auto Memory dry-runs each patch and discards any that do not
+    apply cleanly.
+5.  **Notification.** When a run produces new skills or patches, Gemini CLI
+    surfaces an inline message telling you how many items are waiting.
+
+## How to review extracted skills
+
+Use the `/memory inbox` slash command to open the inbox dialog at any time:
+
+**Command:** `/memory inbox`
+
+The dialog lists each draft skill with its name, description, and source
+sessions. From there you can:
+
+- **Read** the full `SKILL.md` body before deciding.
+- **Promote** a skill to your user (`~/.gemini/skills/`) or workspace
+  (`.gemini/skills/`) directory.
+- **Discard** a skill you do not want.
+- **Apply** or reject a `.patch` proposal against an existing skill.
+
+Promoted skills become discoverable in the next session and follow the standard
+[skill discovery precedence](./skills.md#skill-discovery-tiers).
+
+## How to disable Auto Memory
+
+To turn off background extraction, set the flag back to `false` in your settings
+file and restart Gemini CLI:
+
+```json
+{
+  "experimental": {
+    "autoMemory": false
+  }
+}
+```
+
+Disabling the flag stops the background service immediately on the next session
+start. Existing inbox items remain on disk; you can either drain them with
+`/memory inbox` first or remove the project memory directory manually.
+
+## Data and privacy
+
+- Auto Memory only reads session files that already exist locally on your
+  machine. Nothing is uploaded to Gemini outside the normal API calls the
+  extraction sub-agent makes during its run.
+- The sub-agent is instructed to redact secrets, tokens, and credentials it
+  encounters and to never copy large tool outputs verbatim.
+- Drafted skills live in your project's memory directory until you promote or
+  discard them. They are not automatically loaded into any session.
+
+## Limitations
+
+- The sub-agent runs on a preview Gemini Flash model. Extraction quality depends
+  on the model's ability to recognize durable patterns versus one-off incidents.
+- Auto Memory does not extract skills from the current session. It only
+  considers sessions that have been idle for three hours or more.
+- Inbox items are stored per project. Skills extracted in one workspace are not
+  visible from another until you promote them to the user-scope skills
+  directory.
+
+## Next steps
+
+- Learn how skills are discovered and activated in [Agent Skills](./skills.md).
+- Explore the [memory management tutorial](./tutorials/memory-management.md) for
+  the complementary `save_memory` and `GEMINI.md` workflows.
+- Review the experimental settings catalog in
+  [Settings](./settings.md#experimental).

docs/cli/cli-reference.md

@@ -52,6 +52,7 @@ These commands are available within the interactive REPL.
 | `--prompt-interactive`           | `-i`  | string  | -         | Execute prompt and continue in interactive mode                                                                                                                        |
 | `--worktree`                     | `-w`  | string  | -         | Start Gemini in a new git worktree. If no name is provided, one is generated automatically. Requires `experimental.worktrees: true` in settings.                       |
 | `--sandbox`                      | `-s`  | boolean | `false`   | Run in a sandboxed environment for safer execution                                                                                                                     |
+| `--skip-trust`                   | -     | boolean | `false`   | Trust the current workspace for this session, skipping the folder trust check.                                                                                         |
 | `--approval-mode`                | -     | string  | `default` | Approval mode for tool execution. Choices: `default`, `auto_edit`, `yolo`, `plan`                                                                                      |
 | `--yolo`                         | `-y`  | boolean | `false`   | **Deprecated.** Auto-approve all actions. Use `--approval-mode=yolo` instead.                                                                                          |
 | `--experimental-acp`             | -     | boolean | -         | Start in ACP (Agent Code Pilot) mode. **Experimental feature.**                                                                                                        |

docs/cli/enterprise.md

@@ -507,7 +507,7 @@ events. For more information, see the [telemetry documentation](./telemetry.md).
 You can enforce a specific authentication method for all users by setting the
 `security.auth.enforcedType` in the system-level `settings.json` file. This
 prevents users from choosing a different authentication method. See the
-[Authentication docs](../get-started/authentication.md) for more details.
+[Authentication docs](../get-started/authentication.mdx) for more details.
 
 **Example:** Enforce the use of Google login for all users.
 

docs/cli/plan-mode.md

@@ -130,7 +130,9 @@ These are the only allowed tools:
   [`cli_help`](../core/subagents.md#cli-help-agent)
 - **Interaction:** [`ask_user`](../tools/ask-user.md)
 - **MCP tools (Read):** Read-only [MCP tools](../tools/mcp-server.md) (for
-  example, `github_read_issue`, `postgres_read_schema`) are allowed.
+  example, `github_read_issue`, `postgres_read_schema`) and core
+  [MCP resource tools](../tools/mcp-resources.md) (`list_mcp_resources`,
+  `read_mcp_resource`) are allowed.
 - **Planning (Write):**
   [`write_file`](../tools/file-system.md#3-write_file-writefile) and
   [`replace`](../tools/file-system.md#6-replace-edit) only allowed for `.md`
@@ -329,7 +331,6 @@ Storage whenever Gemini CLI exits Plan Mode to start the implementation.
 #!/usr/bin/env bash
 # Extract the plan filename from the tool input JSON
 plan_filename=$(jq -r '.tool_input.plan_filename // empty')
-plan_filename=$(basename -- "$plan_filename")
 
 # Construct the absolute path using the GEMINI_PLANS_DIR environment variable
 plan_path="$GEMINI_PLANS_DIR/$plan_filename"
@@ -358,7 +359,7 @@ To register this `AfterTool` hook, add it to your `settings.json`:
           {
             "name": "archive-plan",
             "type": "command",
-            "command": "./.gemini/hooks/archive-plan.sh"
+            "command": "~/.gemini/hooks/archive-plan.sh"
           }
         ]
       }

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

@@ -24,20 +24,22 @@ they appear in the UI.
 
 ### General
 
-| UI Label                | Setting                            | Description                                                                                                                                                                                                                                                   | Default     |
-| ----------------------- | ---------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------- |
-| Vim Mode                | `general.vimMode`                  | Enable Vim keybindings                                                                                                                                                                                                                                        | `false`     |
-| Default Approval Mode   | `general.defaultApprovalMode`      | The default approval mode for tool execution. 'default' prompts for approval, 'auto_edit' auto-approves edit tools, and 'plan' is read-only mode. YOLO mode (auto-approve all actions) can only be enabled via command line (--yolo or --approval-mode=yolo). | `"default"` |
-| Enable Auto Update      | `general.enableAutoUpdate`         | Enable automatic updates.                                                                                                                                                                                                                                     | `true`      |
-| Enable Notifications    | `general.enableNotifications`      | Enable run-event notifications for action-required prompts and session completion.                                                                                                                                                                            | `false`     |
-| Enable Plan Mode        | `general.plan.enabled`             | Enable Plan Mode for read-only safety during planning.                                                                                                                                                                                                        | `true`      |
-| Plan Directory          | `general.plan.directory`           | The directory where planning artifacts are stored. If not specified, defaults to the system temporary directory. A custom directory requires a policy to allow write access in Plan Mode.                                                                     | `undefined` |
-| Plan Model Routing      | `general.plan.modelRouting`        | Automatically switch between Pro and Flash models based on Plan Mode status. Uses Pro for the planning phase and Flash for the implementation phase.                                                                                                          | `true`      |
-| Retry Fetch Errors      | `general.retryFetchErrors`         | Retry on "exception TypeError: fetch failed sending request" errors.                                                                                                                                                                                          | `true`      |
-| Max Chat Model Attempts | `general.maxAttempts`              | Maximum number of attempts for requests to the main chat model. Cannot exceed 10.                                                                                                                                                                             | `10`        |
-| Debug Keystroke Logging | `general.debugKeystrokeLogging`    | Enable debug logging of keystrokes to the console.                                                                                                                                                                                                            | `false`     |
-| Enable Session Cleanup  | `general.sessionRetention.enabled` | Enable automatic session cleanup                                                                                                                                                                                                                              | `true`      |
-| Keep chat history       | `general.sessionRetention.maxAge`  | Automatically delete chats older than this time period (e.g., "30d", "7d", "24h", "1w")                                                                                                                                                                       | `"30d"`     |
+| UI Label                      | Setting                            | Description                                                                                                                                                                                                                                                   | Default     |
+| ----------------------------- | ---------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------- |
+| Vim Mode                      | `general.vimMode`                  | Enable Vim keybindings                                                                                                                                                                                                                                        | `false`     |
+| Default Approval Mode         | `general.defaultApprovalMode`      | The default approval mode for tool execution. 'default' prompts for approval, 'auto_edit' auto-approves edit tools, and 'plan' is read-only mode. YOLO mode (auto-approve all actions) can only be enabled via command line (--yolo or --approval-mode=yolo). | `"default"` |
+| Enable Auto Update            | `general.enableAutoUpdate`         | Enable automatic updates.                                                                                                                                                                                                                                     | `true`      |
+| Enable Terminal Notifications | `general.enableNotifications`      | Enable terminal run-event notifications for action-required prompts and session completion.                                                                                                                                                                   | `false`     |
+| Terminal Notification Method  | `general.notificationMethod`       | How to send terminal notifications.                                                                                                                                                                                                                           | `"auto"`    |
+| Enable Plan Mode              | `general.plan.enabled`             | Enable Plan Mode for read-only safety during planning.                                                                                                                                                                                                        | `true`      |
+| Plan Directory                | `general.plan.directory`           | The directory where planning artifacts are stored. If not specified, defaults to the system temporary directory. A custom directory requires a policy to allow write access in Plan Mode.                                                                     | `undefined` |
+| Plan Model Routing            | `general.plan.modelRouting`        | Automatically switch between Pro and Flash models based on Plan Mode status. Uses Pro for the planning phase and Flash for the implementation phase.                                                                                                          | `true`      |
+| Retry Fetch Errors            | `general.retryFetchErrors`         | Retry on "exception TypeError: fetch failed sending request" errors.                                                                                                                                                                                          | `true`      |
+| Max Chat Model Attempts       | `general.maxAttempts`              | Maximum number of attempts for requests to the main chat model. Cannot exceed 10.                                                                                                                                                                             | `10`        |
+| Debug Keystroke Logging       | `general.debugKeystrokeLogging`    | Enable debug logging of keystrokes to the console.                                                                                                                                                                                                            | `false`     |
+| Enable Session Cleanup        | `general.sessionRetention.enabled` | Enable automatic session cleanup                                                                                                                                                                                                                              | `true`      |
+| Keep chat history             | `general.sessionRetention.maxAge`  | Automatically delete chats older than this time period (e.g., "30d", "7d", "24h", "1w")                                                                                                                                                                       | `"30d"`     |
+| Topic & Update Narration      | `general.topicUpdateNarration`     | Enable the Topic & Update communication model for reduced chattiness and structured progress reporting.                                                                                                                                                       | `true`      |
 
 ### Output
 
@@ -159,17 +161,25 @@ they appear in the UI.
 
 ### Experimental
 
-| UI Label                                             | Setting                             | Description                                                                                                                                               | Default |
-| ---------------------------------------------------- | ----------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------- | ------- |
-| 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` |
-| Memory Manager Agent                                 | `experimental.memoryManager`        | Replace the built-in save_memory tool with a memory manager subagent that supports adding, removing, de-duplicating, and organizing memories.             | `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` |
-| Topic & Update Narration                             | `experimental.topicUpdateNarration` | Enable the experimental Topic & Update communication model for reduced chattiness and structured progress reporting.                                      | `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/system-prompt.md

@@ -24,7 +24,7 @@ project-specific behavior or create a customized persona.
 
 You can set the environment variable temporarily in your shell, or persist it
 via a `.gemini/.env` file. See
-[Persisting Environment Variables](../get-started/authentication.md#persisting-environment-variables).
+[Persisting Environment Variables](../get-started/authentication.mdx#persisting-environment-variables).
 
 - Use the project default path (`.gemini/system.md`):
   - `GEMINI_SYSTEM_MD=true` or `GEMINI_SYSTEM_MD=1`
@@ -51,7 +51,7 @@ error with: `missing system prompt file '<path>'`.
   - Create `.gemini/system.md`, then add to `.gemini/.env`:
     - `GEMINI_SYSTEM_MD=1`
 - Use a custom file under your home directory:
-  - `GEMINI_SYSTEM_MD=~/prompts/SYSTEM.md gemini`
+  - `GEMINI_SYSTEM_MD=~/prompts/system.md gemini`
 
 ## UI indicator
 
@@ -102,17 +102,17 @@ safety and workflow rules.
 
 This creates the file and writes the current built‑in system prompt to it.
 
-## Best practices: SYSTEM.md vs GEMINI.md
+## Best practices: system.md vs GEMINI.md
 
-- SYSTEM.md (firmware):
+- system.md (firmware):
   - Non‑negotiable operational rules: safety, tool‑use protocols, approvals, and
     mechanics that keep the CLI reliable.
   - Stable across tasks and projects (or per project when needed).
 - GEMINI.md (strategy):
   - Persona, goals, methodologies, and project/domain context.
-  - Evolves per task; relies on SYSTEM.md for safe execution.
+  - Evolves per task; relies on system.md for safe execution.
 
-Keep SYSTEM.md minimal but complete for safety and tool operation. Keep
+Keep system.md minimal but complete for safety and tool operation. Keep
 GEMINI.md focused on high‑level guidance and project specifics.
 
 ## Troubleshooting

docs/cli/telemetry.md

@@ -35,17 +35,18 @@ The observability system provides:
 You control telemetry behavior through the `.gemini/settings.json` file.
 Environment variables can override these settings.
 
-| Setting        | Environment Variable             | Description                                         | Values            | Default                 |
-| -------------- | -------------------------------- | --------------------------------------------------- | ----------------- | ----------------------- |
-| `enabled`      | `GEMINI_TELEMETRY_ENABLED`       | Enable or disable telemetry                         | `true`/`false`    | `false`                 |
-| `target`       | `GEMINI_TELEMETRY_TARGET`        | Where to send telemetry data                        | `"gcp"`/`"local"` | `"local"`               |
-| `otlpEndpoint` | `GEMINI_TELEMETRY_OTLP_ENDPOINT` | OTLP collector endpoint                             | URL string        | `http://localhost:4317` |
-| `otlpProtocol` | `GEMINI_TELEMETRY_OTLP_PROTOCOL` | OTLP transport protocol                             | `"grpc"`/`"http"` | `"grpc"`                |
-| `outfile`      | `GEMINI_TELEMETRY_OUTFILE`       | Save telemetry to file (overrides `otlpEndpoint`)   | file path         | -                       |
-| `logPrompts`   | `GEMINI_TELEMETRY_LOG_PROMPTS`   | Include prompts in telemetry logs                   | `true`/`false`    | `true`                  |
-| `useCollector` | `GEMINI_TELEMETRY_USE_COLLECTOR` | Use external OTLP collector (advanced)              | `true`/`false`    | `false`                 |
-| `useCliAuth`   | `GEMINI_TELEMETRY_USE_CLI_AUTH`  | Use CLI credentials for telemetry (GCP target only) | `true`/`false`    | `false`                 |
-| -              | `GEMINI_CLI_SURFACE`             | Optional custom label for traffic reporting         | string            | -                       |
+| Setting        | Environment Variable              | Description                                         | Values            | Default                 |
+| -------------- | --------------------------------- | --------------------------------------------------- | ----------------- | ----------------------- |
+| `enabled`      | `GEMINI_TELEMETRY_ENABLED`        | Enable or disable telemetry                         | `true`/`false`    | `false`                 |
+| `traces`       | `GEMINI_TELEMETRY_TRACES_ENABLED` | Enable detailed attribute tracing                   | `true`/`false`    | `false`                 |
+| `target`       | `GEMINI_TELEMETRY_TARGET`         | Where to send telemetry data                        | `"gcp"`/`"local"` | `"local"`               |
+| `otlpEndpoint` | `GEMINI_TELEMETRY_OTLP_ENDPOINT`  | OTLP collector endpoint                             | URL string        | `http://localhost:4317` |
+| `otlpProtocol` | `GEMINI_TELEMETRY_OTLP_PROTOCOL`  | OTLP transport protocol                             | `"grpc"`/`"http"` | `"grpc"`                |
+| `outfile`      | `GEMINI_TELEMETRY_OUTFILE`        | Save telemetry to file (overrides `otlpEndpoint`)   | file path         | -                       |
+| `logPrompts`   | `GEMINI_TELEMETRY_LOG_PROMPTS`    | Include prompts in telemetry logs                   | `true`/`false`    | `true`                  |
+| `useCollector` | `GEMINI_TELEMETRY_USE_COLLECTOR`  | Use external OTLP collector (advanced)              | `true`/`false`    | `false`                 |
+| `useCliAuth`   | `GEMINI_TELEMETRY_USE_CLI_AUTH`   | Use CLI credentials for telemetry (GCP target only) | `true`/`false`    | `false`                 |
+| -              | `GEMINI_CLI_SURFACE`              | Optional custom label for traffic reporting         | string            | -                       |
 
 **Note on boolean environment variables:** For boolean settings like `enabled`,
 setting the environment variable to `true` or `1` enables the feature.
@@ -1235,6 +1236,12 @@ These metrics follow standard [OpenTelemetry GenAI semantic conventions].
 Traces provide an "under-the-hood" view of agent and backend operations. Use
 traces to debug tool interactions and optimize performance.
 
+<!-- prettier-ignore -->
+> [!NOTE]
+> Detailed trace attributes (like full prompts and tool outputs) are disabled by default
+> to minimize overhead. You must explicitly set `telemetry.traces` to `true` (or set
+> `GEMINI_TELEMETRY_TRACES_ENABLED=true`) to capture them.
+
 Every trace captures rich metadata via standard span attributes.
 
 <details open>

docs/cli/trusted-folders.md

@@ -100,6 +100,34 @@ protect you. In this mode, the following features are disabled:
 Granting trust to a folder unlocks the full functionality of Gemini CLI for that
 workspace.
 
+## Headless and automated environments
+
+When running Gemini CLI in a headless environment (for example, a CI/CD
+pipeline) where interactive prompts are not possible, the trust dialog cannot be
+displayed. If the folder is untrusted and the Folder Trust feature is enabled,
+the CLI will throw a `FatalUntrustedWorkspaceError` and exit.
+
+To proceed in these environments, you can bypass the trust check using one of
+the following methods:
+
+- **Command-line flag:** Run the CLI with the `--skip-trust` flag.
+- **Environment variable:** Set the `GEMINI_CLI_TRUST_WORKSPACE=true`
+  environment variable.
+
+These methods will trust the current workspace for the duration of the session
+without prompting.
+
+For detailed instructions on managing folder trust within CI/CD workflows,
+review the
+[Gemini CLI trust guidance for GitHub Actions](https://github.com/google-github-actions/run-gemini-cli/blob/main/docs/trust-guidance.md).
+
+## Overriding the trust file location
+
+By default, trust settings are saved to `~/.gemini/trustedFolders.json`. If you
+need to store this file in a different location, you can set the
+`GEMINI_CLI_TRUSTED_FOLDERS_PATH` environment variable to the desired absolute
+file path.
+
 ## Managing your trust settings
 
 If you need to change a decision or see all your settings, you have a couple of

docs/cli/tutorials/memory-management.md

@@ -124,3 +124,5 @@ immediately. Force a reload with:
 - Explore the [Command reference](../../reference/commands.md) for more
   `/memory` options.
 - Read the technical spec for [Project context](../../cli/gemini-md.md).
+- Try the experimental [Auto Memory](../auto-memory.md) feature to extract
+  reusable skills from your past sessions automatically.

docs/get-started/authentication.mdx

@@ -1,10 +1,11 @@
+import { Tabs, TabItem } from '@astrojs/starlight/components';
+
 # Gemini CLI authentication setup
 
 To use Gemini CLI, you'll need to authenticate with Google. This guide helps you
 quickly find the best way to sign in based on your account type and how you're
 using the CLI.
 
-<!-- prettier-ignore -->
 > [!TIP]
 > Looking for a high-level comparison of all available subscriptions?
 > To compare features and find the right quota for your needs, see our
@@ -23,7 +24,7 @@ Select the authentication method that matches your situation in the table below:
 | Organization users with a company, school, or Google Workspace account | [Sign in with Google](#login-google)                             | [Yes](#set-gcp)                                             |
 | AI Studio user with a Gemini API key                                   | [Use Gemini API Key](#gemini-api)                                | No                                                          |
 | Google Cloud Vertex AI user                                            | [Vertex AI](#vertex-ai)                                          | [Yes](#set-gcp)                                             |
-| [Headless mode](#headless)                                             | [Use Gemini API Key](#gemini-api) or<br> [Vertex AI](#vertex-ai) | No (for Gemini API Key)<br> [Yes](#set-gcp) (for Vertex AI) |
+| [Headless mode](#headless)                                             | [Use Gemini API Key](#gemini-api) or<br /> [Vertex AI](#vertex-ai) | No (for Gemini API Key)<br /> [Yes](#set-gcp) (for Vertex AI) |
 
 ### What is my Google account type?
 
@@ -84,19 +85,24 @@ To authenticate and use Gemini CLI with a Gemini API key:
 
 2. Set the `GEMINI_API_KEY` environment variable to your key. For example:
 
-   **macOS/Linux**
+   <Tabs>
+     <TabItem label="macOS/Linux">
 
-   ```bash
-   # Replace YOUR_GEMINI_API_KEY with the key from AI Studio
-   export GEMINI_API_KEY="YOUR_GEMINI_API_KEY"
-   ```
+     ```bash
+      # Replace YOUR_GEMINI_API_KEY with the key from AI Studio
+      export GEMINI_API_KEY="YOUR_GEMINI_API_KEY"
+     ```
 
-   **Windows (PowerShell)**
+     </TabItem>
+     <TabItem label="Windows (PowerShell)">
 
-   ```powershell
-   # Replace YOUR_GEMINI_API_KEY with the key from AI Studio
-   $env:GEMINI_API_KEY="YOUR_GEMINI_API_KEY"
-   ```
+     ```powershell
+      # Replace YOUR_GEMINI_API_KEY with the key from AI Studio
+      $env:GEMINI_API_KEY="YOUR_GEMINI_API_KEY"
+     ```
+
+     </TabItem>
+   </Tabs>
 
    To make this setting persistent, see
    [Persisting Environment Variables](#persisting-vars).
@@ -109,7 +115,6 @@ To authenticate and use Gemini CLI with a Gemini API key:
 
 4. Select **Use Gemini API key**.
 
-<!-- prettier-ignore -->
 > [!WARNING]
 > Treat API keys, especially for services like Gemini, as sensitive
 > credentials. Protect them to prevent unauthorized access and potential misuse
@@ -131,21 +136,26 @@ or the location where you want to run your jobs.
 
 For example:
 
-**macOS/Linux**
+<Tabs>
+  <TabItem label="macOS/Linux">
 
-```bash
+  ```bash
 # Replace with your project ID and desired location (for example, us-central1)
 export GOOGLE_CLOUD_PROJECT="YOUR_PROJECT_ID"
 export GOOGLE_CLOUD_LOCATION="YOUR_PROJECT_LOCATION"
-```
+  ```
 
-**Windows (PowerShell)**
+  </TabItem>
+  <TabItem label="Windows (PowerShell)">
 
-```powershell
+  ```powershell
 # Replace with your project ID and desired location (for example, us-central1)
 $env:GOOGLE_CLOUD_PROJECT="YOUR_PROJECT_ID"
 $env:GOOGLE_CLOUD_LOCATION="YOUR_PROJECT_LOCATION"
-```
+  ```
+
+  </TabItem>
+</Tabs>
 
 To make any Vertex AI environment variable settings persistent, see
 [Persisting Environment Variables](#persisting-vars).
@@ -157,17 +167,22 @@ Consider this authentication method if you have Google Cloud CLI installed.
 If you have previously set `GOOGLE_API_KEY` or `GEMINI_API_KEY`, you must unset
 them to use ADC.
 
-**macOS/Linux**
+<Tabs>
+  <TabItem label="macOS/Linux">
 
-```bash
+  ```bash
 unset GOOGLE_API_KEY GEMINI_API_KEY
-```
+  ```
 
-**Windows (PowerShell)**
+  </TabItem>
+  <TabItem label="Windows (PowerShell)">
 
-```powershell
+  ```powershell
 Remove-Item Env:\GOOGLE_API_KEY, Env:\GEMINI_API_KEY -ErrorAction Ignore
-```
+  ```
+
+  </TabItem>
+</Tabs>
 
 1. Verify you have a Google Cloud project and Vertex AI API is enabled.
 
@@ -195,17 +210,22 @@ pipelines, or if your organization restricts user-based ADC or API key creation.
 If you have previously set `GOOGLE_API_KEY` or `GEMINI_API_KEY`, you must unset
 them:
 
-**macOS/Linux**
+<Tabs>
+  <TabItem label="macOS/Linux">
 
-```bash
+  ```bash
 unset GOOGLE_API_KEY GEMINI_API_KEY
-```
+  ```
 
-**Windows (PowerShell)**
+  </TabItem>
+  <TabItem label="Windows (PowerShell)">
 
-```powershell
+  ```powershell
 Remove-Item Env:\GOOGLE_API_KEY, Env:\GEMINI_API_KEY -ErrorAction Ignore
-```
+  ```
+
+  </TabItem>
+</Tabs>
 
 1.  [Create a service account and key](https://cloud.google.com/iam/docs/keys-create-delete)
     and download the provided JSON file. Assign the "Vertex AI User" role to the
@@ -214,19 +234,24 @@ Remove-Item Env:\GOOGLE_API_KEY, Env:\GEMINI_API_KEY -ErrorAction Ignore
 2.  Set the `GOOGLE_APPLICATION_CREDENTIALS` environment variable to the JSON
     file's absolute path. For example:
 
-    **macOS/Linux**
+    <Tabs>
+      <TabItem label="macOS/Linux">
 
-    ```bash
-    # Replace /path/to/your/keyfile.json with the actual path
-    export GOOGLE_APPLICATION_CREDENTIALS="/path/to/your/keyfile.json"
-    ```
+      ```bash
+        # Replace /path/to/your/keyfile.json with the actual path
+        export GOOGLE_APPLICATION_CREDENTIALS="/path/to/your/keyfile.json"
+      ```
 
-    **Windows (PowerShell)**
+      </TabItem>
+      <TabItem label="Windows (PowerShell)">
 
-    ```powershell
-    # Replace C:\path\to\your\keyfile.json with the actual path
-    $env:GOOGLE_APPLICATION_CREDENTIALS="C:\path\to\your\keyfile.json"
-    ```
+      ```powershell
+        # Replace C:\path\to\your\keyfile.json with the actual path
+        $env:GOOGLE_APPLICATION_CREDENTIALS="C:\path\to\your\keyfile.json"
+      ```
+
+      </TabItem>
+    </Tabs>
 
 3.  [Configure your Google Cloud Project](#set-gcp).
 
@@ -238,7 +263,6 @@ Remove-Item Env:\GOOGLE_API_KEY, Env:\GEMINI_API_KEY -ErrorAction Ignore
 
 5.  Select **Vertex AI**.
 
-<!-- prettier-ignore -->
 > [!WARNING]
 > Protect your service account key file as it gives access to
 > your resources.
@@ -250,19 +274,24 @@ Remove-Item Env:\GOOGLE_API_KEY, Env:\GEMINI_API_KEY -ErrorAction Ignore
 
 2.  Set the `GOOGLE_API_KEY` environment variable:
 
-    **macOS/Linux**
+    <Tabs>
+      <TabItem label="macOS/Linux">
 
-    ```bash
-    # Replace YOUR_GOOGLE_API_KEY with your Vertex AI API key
-    export GOOGLE_API_KEY="YOUR_GOOGLE_API_KEY"
-    ```
+      ```bash
+        # Replace YOUR_GOOGLE_API_KEY with your Vertex AI API key
+        export GOOGLE_API_KEY="YOUR_GOOGLE_API_KEY"
+      ```
 
-    **Windows (PowerShell)**
+      </TabItem>
+      <TabItem label="Windows (PowerShell)">
 
-    ```powershell
-    # Replace YOUR_GOOGLE_API_KEY with your Vertex AI API key
-    $env:GOOGLE_API_KEY="YOUR_GOOGLE_API_KEY"
-    ```
+      ```powershell
+        # Replace YOUR_GOOGLE_API_KEY with your Vertex AI API key
+        $env:GOOGLE_API_KEY="YOUR_GOOGLE_API_KEY"
+      ```
+
+      </TabItem>
+    </Tabs>
 
     If you see errors like `"API keys are not supported by this API..."`, your
     organization might restrict API key usage for this service. Try the other
@@ -280,7 +309,6 @@ Remove-Item Env:\GOOGLE_API_KEY, Env:\GEMINI_API_KEY -ErrorAction Ignore
 
 ## Set your Google Cloud project <a id="set-gcp"></a>
 
-<!-- prettier-ignore -->
 > [!IMPORTANT]
 > Most individual Google accounts (free and paid) don't require a
 > Google Cloud project for authentication.
@@ -308,19 +336,24 @@ To configure Gemini CLI to use a Google Cloud project, do the following:
 
     For example, to set the `GOOGLE_CLOUD_PROJECT_ID` variable:
 
-    **macOS/Linux**
+    <Tabs>
+      <TabItem label="macOS/Linux">
 
-    ```bash
-    # Replace YOUR_PROJECT_ID with your actual Google Cloud project ID
-    export GOOGLE_CLOUD_PROJECT="YOUR_PROJECT_ID"
-    ```
+      ```bash
+        # Replace YOUR_PROJECT_ID with your actual Google Cloud project ID
+        export GOOGLE_CLOUD_PROJECT="YOUR_PROJECT_ID"
+      ```
 
-    **Windows (PowerShell)**
+      </TabItem>
+      <TabItem label="Windows (PowerShell)">
 
-    ```powershell
-    # Replace YOUR_PROJECT_ID with your actual Google Cloud project ID
-    $env:GOOGLE_CLOUD_PROJECT="YOUR_PROJECT_ID"
-    ```
+      ```powershell
+        # Replace YOUR_PROJECT_ID with your actual Google Cloud project ID
+        $env:GOOGLE_CLOUD_PROJECT="YOUR_PROJECT_ID"
+      ```
+
+      </TabItem>
+    </Tabs>
 
     To make this setting persistent, see
     [Persisting Environment Variables](#persisting-vars).
@@ -333,21 +366,29 @@ persist them with the following methods:
 1.  **Add your environment variables to your shell configuration file:** Append
     the environment variable commands to your shell's startup file.
 
-    **macOS/Linux** (for example, `~/.bashrc`, `~/.zshrc`, or `~/.profile`):
+    <Tabs>
+      <TabItem label="macOS/Linux">
 
-    ```bash
-    echo 'export GOOGLE_CLOUD_PROJECT="YOUR_PROJECT_ID"' >> ~/.bashrc
-    source ~/.bashrc
-    ```
+      (for example, `~/.bashrc`, `~/.zshrc`, or `~/.profile`):
 
-    **Windows (PowerShell)** (for example, `$PROFILE`):
+      ```bash
+        echo 'export GOOGLE_CLOUD_PROJECT="YOUR_PROJECT_ID"' >> ~/.bashrc
+        source ~/.bashrc
+      ```
 
-    ```powershell
-    Add-Content -Path $PROFILE -Value '$env:GOOGLE_CLOUD_PROJECT="YOUR_PROJECT_ID"'
-    . $PROFILE
-    ```
+      </TabItem>
+      <TabItem label="Windows (PowerShell)">
+
+      (for example, `$PROFILE`):
+
+      ```powershell
+        Add-Content -Path $PROFILE -Value '$env:GOOGLE_CLOUD_PROJECT="YOUR_PROJECT_ID"'
+        . $PROFILE
+      ```
+
+      </TabItem>
+    </Tabs>
 
-<!-- prettier-ignore -->
 > [!WARNING]
 > Be aware that when you export API keys or service account
 > paths in your shell configuration file, any process launched from that
@@ -361,25 +402,30 @@ persist them with the following methods:
 
     Example for user-wide settings:
 
-    **macOS/Linux**
+    <Tabs>
+      <TabItem label="macOS/Linux">
 
-    ```bash
-    mkdir -p ~/.gemini
-    cat >> ~/.gemini/.env <<'EOF'
-    GOOGLE_CLOUD_PROJECT="your-project-id"
-    # Add other variables like GEMINI_API_KEY as needed
-    EOF
-    ```
+      ```bash
+        mkdir -p ~/.gemini
+        cat >> ~/.gemini/.env <<'EOF'
+        GOOGLE_CLOUD_PROJECT="your-project-id"
+        # Add other variables like GEMINI_API_KEY as needed
+        EOF
+      ```
 
-    **Windows (PowerShell)**
+      </TabItem>
+      <TabItem label="Windows (PowerShell)">
 
-    ```powershell
-    New-Item -ItemType Directory -Force -Path "$env:USERPROFILE\.gemini"
-    @"
-    GOOGLE_CLOUD_PROJECT="your-project-id"
-    # Add other variables like GEMINI_API_KEY as needed
-    "@ | Out-File -FilePath "$env:USERPROFILE\.gemini\.env" -Encoding utf8 -Append
-    ```
+      ```powershell
+        New-Item -ItemType Directory -Force -Path "$env:USERPROFILE\.gemini"
+        @"
+        GOOGLE_CLOUD_PROJECT="your-project-id"
+        # Add other variables like GEMINI_API_KEY as needed
+        "@ | Out-File -FilePath "$env:USERPROFILE\.gemini\.env" -Encoding utf8 -Append
+      ```
+
+      </TabItem>
+    </Tabs>
 
 Variables are loaded from the first file found, not merged.
 

docs/get-started/index.md

@@ -24,7 +24,8 @@ Once Gemini CLI is installed, run Gemini CLI from your command line:
 gemini
 ```
 
-For more installation options, see [Gemini CLI Installation](./installation.md).
+For more installation options, see
+[Gemini CLI Installation](./installation.mdx).
 
 ## Authenticate
 
@@ -46,7 +47,7 @@ cases, you can log in with your existing Google account:
 
 Certain account types may require you to configure a Google Cloud project. For
 more information, including other authentication methods, see
-[Gemini CLI Authentication Setup](./authentication.md).
+[Gemini CLI Authentication Setup](./authentication.mdx).
 
 ## Configure
 

docs/get-started/installation.md

@@ -1,181 +0,0 @@
-# Gemini CLI installation, execution, and releases
-
-This document provides an overview of Gemini CLI's system requirements,
-installation methods, and release types.
-
-## Recommended system specifications
-
-- **Operating System:**
-  - macOS 15+
-  - Windows 11 24H2+
-  - Ubuntu 20.04+
-- **Hardware:**
-  - "Casual" usage: 4GB+ RAM (short sessions, common tasks and edits)
-  - "Power" usage: 16GB+ RAM (long sessions, large codebases, deep context)
-- **Runtime:** Node.js 20.0.0+
-- **Shell:** Bash, Zsh, or PowerShell
-- **Location:**
-  [Gemini Code Assist supported locations](https://developers.google.com/gemini-code-assist/resources/available-locations#americas)
-- **Internet connection required**
-
-## Install Gemini CLI
-
-We recommend most users install Gemini CLI using one of the following
-installation methods:
-
-- npm
-- Homebrew
-- MacPorts
-- Anaconda
-
-Note that Gemini CLI comes pre-installed on
-[**Cloud Shell**](https://docs.cloud.google.com/shell/docs) and
-[**Cloud Workstations**](https://cloud.google.com/workstations).
-
-### Install globally with npm
-
-```bash
-npm install -g @google/gemini-cli
-```
-
-### Install globally with Homebrew (macOS/Linux)
-
-```bash
-brew install gemini-cli
-```
-
-### Install globally with MacPorts (macOS)
-
-```bash
-sudo port install gemini-cli
-```
-
-### Install with Anaconda (for restricted environments)
-
-```bash
-# Create and activate a new environment
-conda create -y -n gemini_env -c conda-forge nodejs
-conda activate gemini_env
-
-# Install Gemini CLI globally via npm (inside the environment)
-npm install -g @google/gemini-cli
-```
-
-## Run Gemini CLI
-
-For most users, we recommend running Gemini CLI with the `gemini` command:
-
-```bash
-gemini
-```
-
-For a list of options and additional commands, see the
-[CLI cheatsheet](../cli/cli-reference.md).
-
-You can also run Gemini CLI using one of the following advanced methods:
-
-- Run instantly with npx. You can run Gemini CLI without permanent installation.
-- In a sandbox. This method offers increased security and isolation.
-- From the source. This is recommended for contributors to the project.
-
-### Run instantly with npx
-
-```bash
-# Using npx (no installation required)
-npx @google/gemini-cli
-```
-
-You can also execute the CLI directly from the main branch on GitHub, which is
-helpful for testing features still in development:
-
-```bash
-npx https://github.com/google-gemini/gemini-cli
-```
-
-### Run in a sandbox (Docker/Podman)
-
-For security and isolation, Gemini CLI can be run inside a container. This is
-the default way that the CLI executes tools that might have side effects.
-
-- **Directly from the registry:** You can run the published sandbox image
-  directly. This is useful for environments where you only have Docker and want
-  to run the CLI.
-  ```bash
-  # Run the published sandbox image
-  docker run --rm -it us-docker.pkg.dev/gemini-code-dev/gemini-cli/sandbox:0.1.1
-  ```
-- **Using the `--sandbox` flag:** If you have Gemini CLI installed locally
-  (using the standard installation described above), you can instruct it to run
-  inside the sandbox container.
-  ```bash
-  gemini --sandbox -y -p "your prompt here"
-  ```
-
-### Run from source (recommended for Gemini CLI contributors)
-
-Contributors to the project will want to run the CLI directly from the source
-code.
-
-- **Development mode:** This method provides hot-reloading and is useful for
-  active development.
-  ```bash
-  # From the root of the repository
-  npm run start
-  ```
-- **Production mode (React optimizations):** This method runs the CLI with React
-  production mode enabled, which is useful for testing performance without
-  development overhead.
-  ```bash
-  # From the root of the repository
-  npm run start:prod
-  ```
-- **Production-like mode (linked package):** This method simulates a global
-  installation by linking your local package. It's useful for testing a local
-  build in a production workflow.
-
-  ```bash
-  # Link the local cli package to your global node_modules
-  npm link packages/cli
-
-  # Now you can run your local version using the `gemini` command
-  gemini
-  ```
-
-## Releases
-
-Gemini CLI has three release channels: nightly, preview, and stable. For most
-users, we recommend the stable release, which is the default installation.
-
-### Stable
-
-New stable releases are published each week. The stable release is the promotion
-of last week's `preview` release along with any bug fixes. The stable release
-uses `latest` tag, but omitting the tag also installs the latest stable release
-by default:
-
-```bash
-# Both commands install the latest stable release.
-npm install -g @google/gemini-cli
-npm install -g @google/gemini-cli@latest
-```
-
-### Preview
-
-New preview releases will be published each week. These releases are not fully
-vetted and may contain regressions or other outstanding issues. Try out the
-preview release by using the `preview` tag:
-
-```bash
-npm install -g @google/gemini-cli@preview
-```
-
-### Nightly
-
-Nightly releases are published every day. The nightly release includes all
-changes from the main branch at time of release. It should be assumed there are
-pending validations and issues. You can help test the latest changes by
-installing with the `nightly` tag:
-
-```bash
-npm install -g @google/gemini-cli@nightly
-```

docs/get-started/installation.mdx

@@ -0,0 +1,201 @@
+import { Tabs, TabItem } from '@astrojs/starlight/components';
+
+# Gemini CLI installation, execution, and releases
+
+This document provides an overview of Gemini CLI's system requirements,
+installation methods, and release types.
+
+## Recommended system specifications
+
+- **Operating System:**
+  - macOS 15+
+  - Windows 11 24H2+
+  - Ubuntu 20.04+
+- **Hardware:**
+  - "Casual" usage: 4GB+ RAM (short sessions, common tasks and edits)
+  - "Power" usage: 16GB+ RAM (long sessions, large codebases, deep context)
+- **Runtime:** Node.js 20.0.0+
+- **Shell:** Bash, Zsh, or PowerShell
+- **Location:**
+  [Gemini Code Assist supported locations](https://developers.google.com/gemini-code-assist/resources/available-locations#americas)
+- **Internet connection required**
+
+## Install Gemini CLI
+
+We recommend most users install Gemini CLI using one of the following
+installation methods. Note that Gemini CLI comes pre-installed on
+[**Cloud Shell**](https://docs.cloud.google.com/shell/docs) and
+[**Cloud Workstations**](https://cloud.google.com/workstations).
+
+<Tabs>
+  <TabItem label="npm">
+
+  Install globally with npm:
+
+  ```bash
+  npm install -g @google/gemini-cli
+  ```
+
+  </TabItem>
+  <TabItem label="Homebrew (macOS/Linux)">
+
+  Install globally with Homebrew:
+
+  ```bash
+  brew install gemini-cli
+  ```
+
+  </TabItem>
+  <TabItem label="MacPorts (macOS)">
+
+  Install globally with MacPorts:
+
+  ```bash
+  sudo port install gemini-cli
+  ```
+
+  </TabItem>
+  <TabItem label="Anaconda">
+
+  Install with Anaconda (for restricted environments):
+
+  ```bash
+  # Create and activate a new environment
+  conda create -y -n gemini_env -c conda-forge nodejs
+  conda activate gemini_env
+
+  # Install Gemini CLI globally via npm (inside the environment)
+  npm install -g @google/gemini-cli
+  ```
+
+  </TabItem>
+</Tabs>
+
+## Run Gemini CLI
+
+For most users, we recommend running Gemini CLI with the `gemini` command:
+
+```bash
+gemini
+```
+
+For a list of options and additional commands, see the
+[CLI cheatsheet](../cli/cli-reference.md).
+
+You can also run Gemini CLI using one of the following advanced methods:
+
+<Tabs>
+  <TabItem label="npx">
+
+  Run instantly with npx. You can run Gemini CLI without permanent installation.
+
+  ```bash
+  # Using npx (no installation required)
+  npx @google/gemini-cli
+  ```
+
+  You can also execute the CLI directly from the main branch on GitHub, which is
+  helpful for testing features still in development:
+
+  ```bash
+  npx https://github.com/google-gemini/gemini-cli
+  ```
+
+  </TabItem>
+  <TabItem label="Docker/Podman Sandbox">
+
+  For security and isolation, Gemini CLI can be run inside a container. This is
+  the default way that the CLI executes tools that might have side effects.
+
+  - **Directly from the registry:** You can run the published sandbox image
+    directly. This is useful for environments where you only have Docker and want
+    to run the CLI.
+    ```bash
+    # Run the published sandbox image
+    docker run --rm -it us-docker.pkg.dev/gemini-code-dev/gemini-cli/sandbox:0.1.1
+    ```
+  - **Using the `--sandbox` flag:** If you have Gemini CLI installed locally
+    (using the standard installation described above), you can instruct it to run
+    inside the sandbox container.
+    ```bash
+    gemini --sandbox -y -p "your prompt here"
+    ```
+
+  </TabItem>
+  <TabItem label="From source">
+
+  Contributors to the project will want to run the CLI directly from the source
+  code.
+
+  - **Development mode:** This method provides hot-reloading and is useful for
+    active development.
+    ```bash
+    # From the root of the repository
+    npm run start
+    ```
+  - **Production mode (React optimizations):** This method runs the CLI with React
+    production mode enabled, which is useful for testing performance without
+    development overhead.
+    ```bash
+    # From the root of the repository
+    npm run start:prod
+    ```
+  - **Production-like mode (linked package):** This method simulates a global
+    installation by linking your local package. It's useful for testing a local
+    build in a production workflow.
+
+    ```bash
+    # Link the local cli package to your global node_modules
+    npm link packages/cli
+
+    # Now you can run your local version using the `gemini` command
+    gemini
+    ```
+
+  </TabItem>
+</Tabs>
+
+## Releases
+
+Gemini CLI has three release channels: stable, preview, and nightly. For most
+users, we recommend the stable release, which is the default installation.
+
+<Tabs>
+  <TabItem label="Stable">
+
+  Stable releases are published each week. A stable release is created from the
+  previous week's preview release along with any bug fixes. The stable release
+  uses the `latest` tag. Omitting the tag also installs the latest stable
+  release by default.
+
+  ```bash
+  # Both commands install the latest stable release.
+  npm install -g @google/gemini-cli
+  npm install -g @google/gemini-cli@latest
+  ```
+
+  </TabItem>
+  <TabItem label="Preview">
+
+  New preview releases will be published each week. These releases are not fully
+  vetted and may contain regressions or other outstanding issues. Try out the
+  preview release by using the `preview` tag:
+
+  ```bash
+  npm install -g @google/gemini-cli@preview
+  ```
+
+  </TabItem>
+  <TabItem label="Nightly">
+
+  Nightly releases are published every day. The nightly release includes all
+  changes from the main branch at time of release. It should be assumed there are
+  pending validations and issues. You can help test the latest changes by
+  installing with the `nightly` tag:
+
+  ```bash
+  npm install -g @google/gemini-cli@nightly
+  ```
+
+  </TabItem>
+</Tabs>

docs/index.md

@@ -15,9 +15,9 @@ npm install -g @google/gemini-cli
 Jump in to Gemini CLI.
 
 - **[Quickstart](./get-started/index.md):** Your first session with Gemini CLI.
-- **[Installation](./get-started/installation.md):** How to install Gemini CLI
+- **[Installation](./get-started/installation.mdx):** How to install Gemini CLI
   on your system.
-- **[Authentication](./get-started/authentication.md):** Setup instructions for
+- **[Authentication](./get-started/authentication.mdx):** Setup instructions for
   personal and enterprise accounts.
 - **[CLI cheatsheet](./cli/cli-reference.md):** A quick reference for common
   commands and options.

docs/reference/configuration.md

@@ -134,10 +134,15 @@ their corresponding top-level category object in your `settings.json` file.
   - **Default:** `true`
 
 - **`general.enableNotifications`** (boolean):
-  - **Description:** Enable run-event notifications for action-required prompts
-    and session completion.
+  - **Description:** Enable terminal run-event notifications for action-required
+    prompts and session completion.
   - **Default:** `false`
 
+- **`general.notificationMethod`** (enum):
+  - **Description:** How to send terminal notifications.
+  - **Default:** `"auto"`
+  - **Values:** `"auto"`, `"osc9"`, `"osc777"`, `"bell"`
+
 - **`general.checkpointing.enabled`** (boolean):
   - **Description:** Enable session checkpointing for recovery
   - **Default:** `false`
@@ -193,6 +198,11 @@ their corresponding top-level category object in your `settings.json` file.
   - **Description:** Minimum retention period (safety limit, defaults to "1d")
   - **Default:** `"1d"`
 
+- **`general.topicUpdateNarration`** (boolean):
+  - **Description:** Enable the Topic & Update communication model for reduced
+    chattiness and structured progress reporting.
+  - **Default:** `true`
+
 #### `output`
 
 - **`output.format`** (enum):
@@ -426,6 +436,20 @@ their corresponding top-level category object in your `settings.json` file.
   - **Default:** `"ask"`
   - **Values:** `"ask"`, `"always"`, `"never"`
 
+- **`billing.vertexAi.requestType`** (enum):
+  - **Description:** Sets the X-Vertex-AI-LLM-Request-Type header for Vertex AI
+    requests.
+  - **Default:** `undefined`
+  - **Values:** `"dedicated"`, `"shared"`
+  - **Requires restart:** Yes
+
+- **`billing.vertexAi.sharedRequestType`** (enum):
+  - **Description:** Sets the X-Vertex-AI-LLM-Shared-Request-Type header for
+    Vertex AI requests.
+  - **Default:** `undefined`
+  - **Values:** `"priority"`, `"flex"`
+  - **Requires restart:** Yes
+
 #### `model`
 
 - **`model.name`** (string):
@@ -539,6 +563,18 @@ their corresponding top-level category object in your `settings.json` file.
           "model": "gemini-2.5-flash-lite"
         }
       },
+      "gemma-4-31b-it": {
+        "extends": "chat-base-3",
+        "modelConfig": {
+          "model": "gemma-4-31b-it"
+        }
+      },
+      "gemma-4-26b-a4b-it": {
+        "extends": "chat-base-3",
+        "modelConfig": {
+          "model": "gemma-4-26b-a4b-it"
+        }
+      },
       "gemini-2.5-flash-base": {
         "extends": "base",
         "modelConfig": {
@@ -810,6 +846,28 @@ their corresponding top-level category object in your `settings.json` file.
           "multimodalToolUse": false
         }
       },
+      "gemma-4-31b-it": {
+        "displayName": "gemma-4-31b-it",
+        "tier": "custom",
+        "family": "gemma-4",
+        "isPreview": false,
+        "isVisible": true,
+        "features": {
+          "thinking": true,
+          "multimodalToolUse": false
+        }
+      },
+      "gemma-4-26b-a4b-it": {
+        "displayName": "gemma-4-26b-a4b-it",
+        "tier": "custom",
+        "family": "gemma-4",
+        "isPreview": false,
+        "isVisible": true,
+        "features": {
+          "thinking": true,
+          "multimodalToolUse": false
+        }
+      },
       "auto": {
         "tier": "auto",
         "isPreview": true,
@@ -880,6 +938,12 @@ their corresponding top-level category object in your `settings.json` file.
 
     ```json
     {
+      "gemma-4-31b-it": {
+        "default": "gemma-4-31b-it"
+      },
+      "gemma-4-26b-a4b-it": {
+        "default": "gemma-4-26b-a4b-it"
+      },
       "gemini-3.1-pro-preview": {
         "default": "gemini-3.1-pro-preview",
         "contexts": [
@@ -1127,7 +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"
           }
@@ -1143,7 +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"
           }
@@ -1160,7 +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"
           }
@@ -1176,7 +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"
           }
@@ -1193,7 +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"
           }
@@ -1208,7 +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"
           }
@@ -1224,7 +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"
           }
@@ -1349,6 +1413,12 @@ their corresponding top-level category object in your `settings.json` file.
   - **Default:** `true`
   - **Requires restart:** Yes
 
+- **`context.fileFiltering.enableFileWatcher`** (boolean):
+  - **Description:** Enable file watcher updates for @ file suggestions
+    (experimental).
+  - **Default:** `false`
+  - **Requires restart:** Yes
+
 - **`context.fileFiltering.enableRecursiveFileSearch`** (boolean):
   - **Description:** Enable recursive file search functionality when completing
     @ references in the prompt.
@@ -1437,6 +1507,12 @@ their corresponding top-level category object in your `settings.json` file.
   - **Default:** `undefined`
   - **Requires restart:** Yes
 
+- **`tools.confirmationRequired`** (array):
+  - **Description:** Tool names that always require user confirmation. Takes
+    precedence over allowed tools and core tool allowlists.
+  - **Default:** `undefined`
+  - **Requires restart:** Yes
+
 - **`tools.exclude`** (array):
   - **Description:** Tool names to exclude from discovery.
   - **Default:** `undefined`
@@ -1610,6 +1686,37 @@ their corresponding top-level category object in your `settings.json` file.
 
 #### `experimental`
 
+- **`experimental.gemma`** (boolean):
+  - **Description:** Enable access to Gemma 4 models (experimental).
+  - **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`
@@ -1658,8 +1765,10 @@ their corresponding top-level category object in your `settings.json` file.
   - **Requires restart:** Yes
 
 - **`experimental.jitContext`** (boolean):
-  - **Description:** Enable Just-In-Time (JIT) context loading.
-  - **Default:** `false`
+  - **Description:** Enable Just-In-Time (JIT) context loading. Defaults to
+    true; set to false to opt out and load all GEMINI.md files into the system
+    instruction up-front.
+  - **Default:** `true`
   - **Requires restart:** Yes
 
 - **`experimental.useOSC52Paste`** (boolean):
@@ -1701,6 +1810,18 @@ their corresponding top-level category object in your `settings.json` file.
   - **Default:** `false`
   - **Requires restart:** Yes
 
+- **`experimental.gemmaModelRouter.autoStartServer`** (boolean):
+  - **Description:** Automatically start the LiteRT-LM server when Gemini CLI
+    starts and the Gemma router is enabled.
+  - **Default:** `false`
+  - **Requires restart:** Yes
+
+- **`experimental.gemmaModelRouter.binaryPath`** (string):
+  - **Description:** Custom path to the LiteRT-LM binary. Leave empty to use the
+    default location (~/.gemini/bin/litert/).
+  - **Default:** `""`
+  - **Requires restart:** Yes
+
 - **`experimental.gemmaModelRouter.classifier.host`** (string):
   - **Description:** The host of the classifier.
   - **Default:** `"http://localhost:9379"`
@@ -1712,10 +1833,28 @@ their corresponding top-level category object in your `settings.json` file.
   - **Default:** `"gemma3-1b-gpu-custom"`
   - **Requires restart:** Yes
 
-- **`experimental.memoryManager`** (boolean):
-  - **Description:** Replace the built-in save_memory tool with a memory manager
-    subagent that supports adding, removing, de-duplicating, and organizing
-    memories.
+- **`experimental.memoryV2`** (boolean):
+  - **Description:** 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.
+  - **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.
   - **Default:** `false`
   - **Requires restart:** Yes
 
@@ -1730,8 +1869,7 @@ their corresponding top-level category object in your `settings.json` file.
   - **Requires restart:** Yes
 
 - **`experimental.topicUpdateNarration`** (boolean):
-  - **Description:** Enable the experimental Topic & Update communication model
-    for reduced chattiness and structured progress reporting.
+  - **Description:** Deprecated: Use general.topicUpdateNarration instead.
   - **Default:** `false`
 
 #### `skills`
@@ -1971,6 +2109,8 @@ see [Telemetry](../cli/telemetry.md).
 
 - **Properties:**
   - **`enabled`** (boolean): Whether or not telemetry is enabled.
+  - **`traces`** (boolean): Whether detailed traces with large attributes (like
+    tool outputs and file reads) are captured. Defaults to `false`.
   - **`target`** (string): The destination for collected telemetry. Supported
     values are `local` and `gcp`.
   - **`otlpEndpoint`** (string): The endpoint for the OTLP Exporter.
@@ -2070,7 +2210,7 @@ within your user's home folder.
 Environment variables are a common way to configure applications, especially for
 sensitive information like API keys or for settings that might change between
 environments. For authentication setup, see the
-[Authentication documentation](../get-started/authentication.md) which covers
+[Authentication documentation](../get-started/authentication.mdx) which covers
 all available authentication methods.
 
 The CLI automatically loads environment variables from an `.env` file. The
@@ -2091,7 +2231,7 @@ the `advanced.excludedEnvVars` setting in your `settings.json` file.
 - **`GEMINI_API_KEY`**:
   - Your API key for the Gemini API.
   - One of several available
-    [authentication methods](../get-started/authentication.md).
+    [authentication methods](../get-started/authentication.mdx).
   - Set this in your shell profile (for example, `~/.bashrc`, `~/.zshrc`) or an
     `.env` file.
 - **`GEMINI_MODEL`**:
@@ -2099,6 +2239,14 @@ the `advanced.excludedEnvVars` setting in your `settings.json` file.
   - Overrides the hardcoded default
   - Example: `export GEMINI_MODEL="gemini-3-flash-preview"` (Windows PowerShell:
     `$env:GEMINI_MODEL="gemini-3-flash-preview"`)
+- **`GEMINI_CLI_TRUST_WORKSPACE`**:
+  - If set to `"true"`, trusts the current workspace for the duration of the
+    session, bypassing the folder trust check.
+  - Useful for headless environments (for example, CI/CD pipelines).
+- **`GEMINI_CLI_TRUSTED_FOLDERS_PATH`**:
+  - Overrides the default location for the `trustedFolders.json` file.
+  - Useful if you want to store this configuration in a custom location instead
+    of the default `~/.gemini/`.
 - **`GEMINI_CLI_IDE_PID`**:
   - Manually specifies the PID of the IDE process to use for integration. This
     is useful when running Gemini CLI in a standalone terminal while still
@@ -2171,6 +2319,10 @@ the `advanced.excludedEnvVars` setting in your `settings.json` file.
   - Set to `true` or `1` to enable telemetry. Any other value is treated as
     disabling it.
   - Overrides the `telemetry.enabled` setting.
+- **`GEMINI_TELEMETRY_TRACES_ENABLED`**:
+  - Set to `true` or `1` to enable detailed tracing with large attributes. Any
+    other value is treated as disabling it.
+  - Overrides the `telemetry.traces` setting.
 - **`GEMINI_TELEMETRY_TARGET`**:
   - Sets the telemetry target (`local` or `gcp`).
   - Overrides the `telemetry.target` setting.

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
 

docs/reference/policy-engine.md

@@ -120,6 +120,12 @@ There are three possible decisions a rule can enforce:
 
 ### Priority system and tiers
 
+> [!WARNING] The **Workspace** tier (project-level policies) is currently
+> non-functional. Defining policies in a workspace's `.gemini/policies`
+> directory will not have any effect. See
+> [issue #18186](https://github.com/google-gemini/gemini-cli/issues/18186). Use
+> User or Admin policies instead.
+
 The policy engine uses a sophisticated priority system to resolve conflicts when
 multiple rules match a single tool call. The core principle is simple: **the
 rule with the highest priority wins**.
@@ -127,13 +133,13 @@ rule with the highest priority wins**.
 To provide a clear hierarchy, policies are organized into three tiers. Each tier
 has a designated number that forms the base of the final priority calculation.
 
-| Tier      | Base | Description                                                                       |
-| :-------- | :--- | :-------------------------------------------------------------------------------- |
-| Default   | 1    | Built-in policies that ship with Gemini CLI.                                      |
-| Extension | 2    | Policies defined in extensions.                                                   |
-| Workspace | 3    | Policies defined in the current workspace's configuration directory.              |
-| User      | 4    | Custom policies defined by the user.                                              |
-| Admin     | 5    | Policies managed by an administrator (for example, in an enterprise environment). |
+| Tier      | Base | Description                                                                                   |
+| :-------- | :--- | :-------------------------------------------------------------------------------------------- |
+| Default   | 1    | Built-in policies that ship with Gemini CLI.                                                  |
+| Extension | 2    | Policies defined in extensions.                                                               |
+| Workspace | 3    | **(Currently disabled)** Policies defined in the current workspace's configuration directory. |
+| User      | 4    | Custom policies defined by the user.                                                          |
+| Admin     | 5    | Policies managed by an administrator (for example, in an enterprise environment).             |
 
 Within a TOML policy file, you assign a priority value from **0 to 999**. The
 engine transforms this into a final priority using the following formula:
@@ -214,11 +220,11 @@ User, and (if configured) Admin directories.
 
 ### Policy locations
 
-| Tier          | Type   | Location                                  |
-| :------------ | :----- | :---------------------------------------- |
-| **User**      | Custom | `~/.gemini/policies/*.toml`               |
-| **Workspace** | Custom | `$WORKSPACE_ROOT/.gemini/policies/*.toml` |
-| **Admin**     | System | _See below (OS specific)_                 |
+| Tier          | Type   | Location                                                 |
+| :------------ | :----- | :------------------------------------------------------- |
+| **User**      | Custom | `~/.gemini/policies/*.toml`                              |
+| **Workspace** | Custom | **(Disabled)** `$WORKSPACE_ROOT/.gemini/policies/*.toml` |
+| **Admin**     | System | _See below (OS specific)_                                |
 
 #### System-wide policies (Admin)
 

docs/reference/tools.md

@@ -92,6 +92,28 @@ each tool.
 | [`ask_user`](../tools/ask-user.md) | `Communicate` | Requests clarification or missing information via an interactive dialog.               |
 | [`write_todos`](../tools/todos.md) | `Other`       | Maintains an internal list of subtasks. The model uses this to track its own progress. |
 
+### Task Tracker (Experimental)
+
+<!-- prettier-ignore -->
+> [!NOTE]
+> This is an experimental feature currently under active development. Enable via `experimental.taskTracker`.
+
+| Tool                                            | Kind    | Description                                                                 |
+| :---------------------------------------------- | :------ | :-------------------------------------------------------------------------- |
+| [`tracker_create_task`](../tools/tracker.md)    | `Other` | Creates a new task in the experimental tracker.                             |
+| [`tracker_update_task`](../tools/tracker.md)    | `Other` | Updates an existing task's status, description, or dependencies.            |
+| [`tracker_get_task`](../tools/tracker.md)       | `Other` | Retrieves the full details of a specific task.                              |
+| [`tracker_list_tasks`](../tools/tracker.md)     | `Other` | Lists tasks in the tracker, optionally filtered by status, type, or parent. |
+| [`tracker_add_dependency`](../tools/tracker.md) | `Other` | Adds a dependency between two tasks, ensuring topological execution.        |
+| [`tracker_visualize`](../tools/tracker.md)      | `Other` | Renders an ASCII tree visualization of the current task graph.              |
+
+### MCP
+
+| Tool                                              | Kind     | Description                                                            |
+| :------------------------------------------------ | :------- | :--------------------------------------------------------------------- |
+| [`list_mcp_resources`](../tools/mcp-resources.md) | `Search` | Lists all available resources exposed by connected MCP servers.        |
+| [`read_mcp_resource`](../tools/mcp-resources.md)  | `Read`   | Reads the content of a specific Model Context Protocol (MCP) resource. |
+
 ### Memory
 
 | Tool                                             | Kind    | Description                                                                          |

docs/sidebar.json

@@ -96,6 +96,11 @@
             ]
           },
           { "label": "Agent Skills", "slug": "docs/cli/skills" },
+          {
+            "label": "Auto Memory",
+            "badge": "🔬",
+            "slug": "docs/cli/auto-memory"
+          },
           { "label": "Checkpointing", "slug": "docs/cli/checkpointing" },
           { "label": "Headless mode", "slug": "docs/cli/headless" },
           {
@@ -122,7 +127,14 @@
               }
             ]
           },
-          { "label": "MCP servers", "slug": "docs/tools/mcp-server" },
+          {
+            "label": "MCP servers",
+            "collapsed": true,
+            "items": [
+              { "label": "Overview", "slug": "docs/tools/mcp-server" },
+              { "label": "Resource tools", "slug": "docs/tools/mcp-resources" }
+            ]
+          },
           { "label": "Model routing", "slug": "docs/cli/model-routing" },
           { "label": "Model selection", "slug": "docs/cli/model" },
           {

docs/tools/mcp-resources.md

@@ -0,0 +1,44 @@
+# MCP resource tools
+
+MCP resource tools let Gemini CLI discover and retrieve data from contextual
+resources exposed by Model Context Protocol (MCP) servers.
+
+## 1. `list_mcp_resources` (ListMcpResources)
+
+`list_mcp_resources` retrieves a list of all available resources from connected
+MCP servers. This is primarily a discovery tool that helps the model understand
+what external data sources are available for reference.
+
+- **Tool name:** `list_mcp_resources`
+- **Display name:** List MCP Resources
+- **Kind:** `Search`
+- **File:** `list-mcp-resources.ts`
+- **Parameters:**
+  - `serverName` (string, optional): An optional filter to list resources from a
+    specific server.
+- **Behavior:**
+  - Iterates through all connected MCP servers.
+  - Fetches the list of resources each server exposes.
+  - Formats the results into a plain-text list of URIs and descriptions.
+- **Output (`llmContent`):** A formatted list of available resources, including
+  their URI, server name, and optional description.
+- **Confirmation:** No. This is a read-only discovery tool.
+
+## 2. `read_mcp_resource` (ReadMcpResource)
+
+`read_mcp_resource` retrieves the content of a specific resource identified by
+its URI.
+
+- **Tool name:** `read_mcp_resource`
+- **Display name:** Read MCP Resource
+- **Kind:** `Read`
+- **File:** `read-mcp-resource.ts`
+- **Parameters:**
+  - `uri` (string, required): The URI of the MCP resource to read.
+- **Behavior:**
+  - Locates the resource and its associated server by URI.
+  - Calls the server's `resources/read` method.
+  - Processes the response, extracting text or binary data.
+- **Output (`llmContent`):** The content of the resource. For binary data, it
+  returns a placeholder indicating the data type.
+- **Confirmation:** No. This is a read-only retrieval tool.

docs/tools/mcp-server.md

@@ -64,7 +64,8 @@ Gemini CLI supports three MCP transport types:
 
 Some MCP servers expose contextual “resources” in addition to the tools and
 prompts. Gemini CLI discovers these automatically and gives you the possibility
-to reference them in the chat.
+to reference them in the chat. For more information on the tools used to
+interact with these resources, see [MCP resource tools](mcp-resources.md).
 
 ### Discovery and listing
 

docs/tools/tracker.md

@@ -0,0 +1,61 @@
+# Tracker tools (`tracker_*`)
+
+<!-- prettier-ignore -->
+> [!NOTE]
+> This is an experimental feature currently under active development.
+
+The `tracker_*` tools allow the Gemini agent to maintain an internal, persistent
+graph of tasks and dependencies for multi-step requests. This suite of tools
+provides a more robust and granular way to manage execution plans than the
+legacy `write_todos` tool.
+
+## Technical reference
+
+The agent uses these tools to manage its execution plan, decompose complex goals
+into actionable sub-tasks, and provide real-time progress updates to the CLI
+interface. The task state is stored in the `.gemini/tmp/tracker/<session-id>`
+directory, allowing the agent to manage its plan for the current session.
+
+### Available Tools
+
+- `tracker_create_task`: Creates a new task in the tracker. You can specify a
+  title, description, and task type (`epic`, `task`, `bug`).
+- `tracker_update_task`: Updates an existing task's status (`open`,
+  `in_progress`, `blocked`, `closed`), description, or dependencies.
+- `tracker_get_task`: Retrieves the full details of a specific task by its
+  6-character hex ID.
+- `tracker_list_tasks`: Lists tasks in the tracker, optionally filtered by
+  status, type, or parent ID.
+- `tracker_add_dependency`: Adds a dependency between two tasks, ensuring
+  topological execution.
+- `tracker_visualize`: Renders an ASCII tree visualization of the current task
+  graph.
+
+## Technical behavior
+
+- **Interface:** Updates the progress indicator and task tree above the CLI
+  input prompt.
+- **Persistence:** Task state is saved automatically to the
+  `.gemini/tmp/tracker/<session-id>` directory. Task states are session-specific
+  and do not persist across different sessions.
+- **Dependencies:** Tasks can depend on other tasks, forming a directed acyclic
+  graph (DAG). The agent must resolve dependencies before starting blocked
+  tasks.
+- **Interaction:** Users can view the current state of the tracker by asking the
+  agent to visualize it, or by running `gemini-cli` commands if implemented.
+
+## Use cases
+
+- Coordinating multi-file refactoring projects.
+- Breaking down a mission into a hierarchy of epics and tasks for better
+  visibility.
+- Tracking bugs and feature requests directly within the context of an active
+  codebase.
+- Providing visibility into the agent's current focus and remaining work.
+
+## Next steps
+
+- Follow the [Task planning tutorial](../cli/tutorials/task-planning.md) for
+  usage details and migration from the legacy todo list.
+- Learn about [Session management](../cli/session-management.md) for context on
+  persistent state.

evals/background_processes.eval.ts

@@ -11,6 +11,8 @@ import path from 'node:path';
 
 describe('Background Process Monitoring', () => {
   evalTest('USUALLY_PASSES', {
+    suiteName: 'default',
+    suiteType: 'behavioral',
     name: 'should naturally use read output tool to find token',
     prompt:
       "Run the script using 'bash generate_token.sh'. It will emit a token after a short delay and continue running. Find the token and tell me what it is.",
@@ -50,6 +52,8 @@ sleep 100
   });
 
   evalTest('USUALLY_PASSES', {
+    suiteName: 'default',
+    suiteType: 'behavioral',
     name: 'should naturally use list tool to verify multiple processes',
     prompt:
       "Start three background processes that run 'sleep 100', 'sleep 200', and 'sleep 300' respectively. Verify that all three are currently running.",

evals/hierarchical_memory.eval.ts

@@ -17,7 +17,7 @@ describe('Hierarchical Memory', () => {
     params: {
       settings: {
         security: {
-          folderTrust: { enabled: true },
+          folderTrust: { enabled: false },
         },
       },
     },
@@ -55,7 +55,7 @@ What is my favorite fruit? Tell me just the name of the fruit.`,
     params: {
       settings: {
         security: {
-          folderTrust: { enabled: true },
+          folderTrust: { enabled: false },
         },
       },
     },
@@ -96,7 +96,7 @@ Provide the answer as an XML block like this:
     params: {
       settings: {
         security: {
-          folderTrust: { enabled: true },
+          folderTrust: { enabled: false },
         },
       },
     },

evals/plan_mode.eval.ts

@@ -298,12 +298,14 @@ describe('plan_mode', () => {
   });
 
   evalTest('ALWAYS_PASSES', {
+    suiteName: 'default',
+    suiteType: 'behavioral',
     name: 'should transition from plan mode to normal execution and create a plan file from scratch',
     params: {
       settings,
     },
     prompt:
-      'Enter plan mode and plan to create a new module called foo. The plan should be saved as foo-plan.md. Then, exit plan mode.',
+      'I agree with your strategy. Please enter plan mode and draft the plan to create a new module called foo. The plan should be saved as foo-plan.md. Then, exit plan mode.',
     assert: async (rig, result) => {
       const enterPlanCalled = await rig.waitForToolCall('enter_plan_mode');
       expect(
@@ -333,7 +335,7 @@ describe('plan_mode', () => {
 
       expect(
         planWrite?.toolRequest.success,
-        `Expected write_file to succeed, but got error: ${planWrite?.toolRequest.error}`,
+        `Expected write_file to succeed, but got error: ${(planWrite?.toolRequest as any).error}`,
       ).toBe(true);
 
       assertModelHasOutput(result);
@@ -341,6 +343,8 @@ describe('plan_mode', () => {
   });
 
   evalTest('USUALLY_PASSES', {
+    suiteName: 'default',
+    suiteType: 'behavioral',
     name: 'should not exit plan mode or draft before informal agreement',
     approvalMode: ApprovalMode.PLAN,
     params: {
@@ -372,4 +376,48 @@ describe('plan_mode', () => {
       assertModelHasOutput(result);
     },
   });
+
+  evalTest('USUALLY_PASSES', {
+    name: 'should handle nested plan directories correctly',
+    suiteName: 'plan_mode',
+    suiteType: 'behavioral',
+    approvalMode: ApprovalMode.PLAN,
+    params: {
+      settings,
+    },
+    prompt:
+      'Please create a new architectural plan in a nested folder called "architecture/frontend-v2.md" within the plans directory. The plan should contain the text "# Frontend V2 Plan". Then, exit plan mode',
+    assert: async (rig, result) => {
+      await rig.waitForTelemetryReady();
+      const toolLogs = rig.readToolLogs();
+
+      const writeCalls = toolLogs.filter((log) =>
+        ['write_file', 'replace'].includes(log.toolRequest.name),
+      );
+
+      const wroteToNestedPath = writeCalls.some((log) => {
+        try {
+          const args = JSON.parse(log.toolRequest.args);
+          if (!args.file_path) return false;
+          // In plan mode, paths can be passed as relative (architecture/frontend-v2.md)
+          // or they might be resolved as absolute by the tool depending on the exact mock state.
+          // We strictly ensure it ends exactly with the expected nested path and doesn't contain extra nesting.
+          const normalizedPath = args.file_path.replace(/\\/g, '/');
+          return (
+            normalizedPath === 'architecture/frontend-v2.md' ||
+            normalizedPath.endsWith('/plans/architecture/frontend-v2.md')
+          );
+        } catch {
+          return false;
+        }
+      });
+
+      expect(
+        wroteToNestedPath,
+        'Expected model to successfully target the nested plan file path',
+      ).toBe(true);
+
+      assertModelHasOutput(result);
+    },
+  });
 });

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";
@@ -18,6 +84,11 @@ describe('save_memory', () => {
     suiteName: 'default',
     suiteType: 'behavioral',
     name: rememberingFavoriteColor,
+    params: {
+      settings: {
+        experimental: { memoryV2: false },
+      },
+    },
 
     prompt: `remember that my favorite color is  blue.
   
@@ -40,6 +111,11 @@ describe('save_memory', () => {
     suiteName: 'default',
     suiteType: 'behavioral',
     name: rememberingCommandRestrictions,
+    params: {
+      settings: {
+        experimental: { memoryV2: false },
+      },
+    },
 
     prompt: `I don't want you to ever run npm commands.`,
     assert: async (rig, result) => {
@@ -61,6 +137,11 @@ describe('save_memory', () => {
     suiteName: 'default',
     suiteType: 'behavioral',
     name: rememberingWorkflow,
+    params: {
+      settings: {
+        experimental: { memoryV2: false },
+      },
+    },
 
     prompt: `I want you to always lint after building.`,
     assert: async (rig, result) => {
@@ -83,6 +164,11 @@ describe('save_memory', () => {
     suiteName: 'default',
     suiteType: 'behavioral',
     name: ignoringTemporaryInformation,
+    params: {
+      settings: {
+        experimental: { memoryV2: false },
+      },
+    },
 
     prompt: `I'm going to get a coffee.`,
     assert: async (rig, result) => {
@@ -108,6 +194,11 @@ describe('save_memory', () => {
     suiteName: 'default',
     suiteType: 'behavioral',
     name: rememberingPetName,
+    params: {
+      settings: {
+        experimental: { memoryV2: false },
+      },
+    },
 
     prompt: `Please remember that my dog's name is Buddy.`,
     assert: async (rig, result) => {
@@ -129,6 +220,11 @@ describe('save_memory', () => {
     suiteName: 'default',
     suiteType: 'behavioral',
     name: rememberingCommandAlias,
+    params: {
+      settings: {
+        experimental: { memoryV2: false },
+      },
+    },
 
     prompt: `When I say 'start server', you should run 'npm run dev'.`,
     assert: async (rig, result) => {
@@ -151,6 +247,11 @@ describe('save_memory', () => {
     suiteName: 'default',
     suiteType: 'behavioral',
     name: savingDbSchemaLocationAsProjectMemory,
+    params: {
+      settings: {
+        experimental: { memoryV2: false },
+      },
+    },
     prompt: `The database schema for this workspace is located in \`db/schema.sql\`.`,
     assert: async (rig, result) => {
       const wasToolCalled = await rig.waitForToolCall(
@@ -180,6 +281,11 @@ describe('save_memory', () => {
     suiteName: 'default',
     suiteType: 'behavioral',
     name: rememberingCodingStyle,
+    params: {
+      settings: {
+        experimental: { memoryV2: false },
+      },
+    },
 
     prompt: `I prefer to use tabs instead of spaces for indentation.`,
     assert: async (rig, result) => {
@@ -202,6 +308,11 @@ describe('save_memory', () => {
     suiteName: 'default',
     suiteType: 'behavioral',
     name: savingBuildArtifactLocationAsProjectMemory,
+    params: {
+      settings: {
+        experimental: { memoryV2: false },
+      },
+    },
     prompt: `In this workspace, build artifacts are stored in the \`dist/artifacts\` directory.`,
     assert: async (rig, result) => {
       const wasToolCalled = await rig.waitForToolCall(
@@ -231,6 +342,11 @@ describe('save_memory', () => {
     suiteName: 'default',
     suiteType: 'behavioral',
     name: savingMainEntryPointAsProjectMemory,
+    params: {
+      settings: {
+        experimental: { memoryV2: false },
+      },
+    },
     prompt: `The main entry point for this workspace is \`src/index.js\`.`,
     assert: async (rig, result) => {
       const wasToolCalled = await rig.waitForToolCall(
@@ -259,6 +375,11 @@ describe('save_memory', () => {
     suiteName: 'default',
     suiteType: 'behavioral',
     name: rememberingBirthday,
+    params: {
+      settings: {
+        experimental: { memoryV2: false },
+      },
+    },
 
     prompt: `My birthday is on June 15th.`,
     assert: async (rig, result) => {
@@ -283,7 +404,7 @@ describe('save_memory', () => {
     name: proactiveMemoryFromLongSession,
     params: {
       settings: {
-        experimental: { memoryManager: true },
+        experimental: { memoryV2: true },
       },
     },
     messages: [
@@ -341,29 +462,75 @@ describe('save_memory', () => {
     prompt:
       'Please save any persistent preferences or facts about me from our conversation to memory.',
     assert: async (rig, result) => {
-      const wasToolCalled = await rig.waitForToolCall(
-        'invoke_agent',
-        undefined,
-        (args) => /save_memory/i.test(args) && /vitest/i.test(args),
-      );
+      // Under experimental.memoryV2, the agent persists memories by
+      // editing markdown files directly with write_file or replace — not via
+      // a save_memory subagent. The user said "I always prefer Vitest over
+      // Jest for testing in all my projects" — that matches the new
+      // cross-project cue phrase ("across all my projects"), so under the
+      // 4-tier model the correct destination is the global personal memory
+      // file (~/.gemini/GEMINI.md). It must NOT land in a committed project
+      // GEMINI.md (that tier is for team conventions) or the per-project
+      // private memory folder (that tier is for project-specific personal
+      // notes). The chat history mixes this durable preference with
+      // transient debugging chatter, so the eval also verifies the agent
+      // picks out the persistent fact among the noise.
+      await rig.waitForToolCall('write_file').catch(() => {});
+      const writeCalls = rig
+        .readToolLogs()
+        .filter((log) =>
+          ['write_file', 'replace'].includes(log.toolRequest.name),
+        );
+
+      const wroteVitestToGlobal = writeCalls.some((log) => {
+        const args = log.toolRequest.args;
+        return (
+          /\.gemini\/GEMINI\.md/i.test(args) &&
+          !/tmp\/[^/]+\/memory/i.test(args) &&
+          /vitest/i.test(args)
+        );
+      });
       expect(
-        wasToolCalled,
-        'Expected invoke_agent to be called with save_memory agent and the Vitest preference from the conversation history',
+        wroteVitestToGlobal,
+        'Expected the cross-project Vitest preference to be written to the global personal memory file (~/.gemini/GEMINI.md) via write_file or replace',
       ).toBe(true);
 
+      const leakedToCommittedProject = writeCalls.some((log) => {
+        const args = log.toolRequest.args;
+        return (
+          /GEMINI\.md/i.test(args) &&
+          !/\.gemini\//i.test(args) &&
+          /vitest/i.test(args)
+        );
+      });
+      expect(
+        leakedToCommittedProject,
+        'Cross-project Vitest preference must NOT be mirrored into a committed project ./GEMINI.md (that tier is for team-shared conventions only)',
+      ).toBe(false);
+
+      const leakedToPrivateProject = writeCalls.some((log) => {
+        const args = log.toolRequest.args;
+        return (
+          /\.gemini\/tmp\/[^/]+\/memory\//i.test(args) && /vitest/i.test(args)
+        );
+      });
+      expect(
+        leakedToPrivateProject,
+        'Cross-project Vitest preference must NOT be mirrored into the private project memory folder (that tier is for project-specific personal notes only)',
+      ).toBe(false);
+
       assertModelHasOutput(result);
     },
   });
 
-  const memoryManagerRoutingPreferences =
-    'Agent routes global and project preferences to memory';
+  const memoryV2RoutesTeamConventionsToProjectGemini =
+    'Agent routes team-shared project conventions to ./GEMINI.md';
   evalTest('USUALLY_PASSES', {
     suiteName: 'default',
     suiteType: 'behavioral',
-    name: memoryManagerRoutingPreferences,
+    name: memoryV2RoutesTeamConventionsToProjectGemini,
     params: {
       settings: {
-        experimental: { memoryManager: true },
+        experimental: { memoryV2: true },
       },
     },
     messages: [
@@ -372,7 +539,7 @@ describe('save_memory', () => {
         type: 'user',
         content: [
           {
-            text: 'I always use dark mode in all my editors and terminals.',
+            text: 'For this project, the team always runs tests with `npm run test` — please remember that as our project convention.',
           },
         ],
         timestamp: '2026-01-01T00:00:00Z',
@@ -380,7 +547,9 @@ describe('save_memory', () => {
       {
         id: 'msg-2',
         type: 'gemini',
-        content: [{ text: 'Got it, I will keep that in mind!' }],
+        content: [
+          { text: 'Got it, I will keep `npm run test` in mind for tests.' },
+        ],
         timestamp: '2026-01-01T00:00:05Z',
       },
       {
@@ -404,15 +573,334 @@ describe('save_memory', () => {
     ],
     prompt: 'Please save the preferences I mentioned earlier to memory.',
     assert: async (rig, result) => {
-      const wasToolCalled = await rig.waitForToolCall(
-        'invoke_agent',
-        undefined,
-        (args) => /save_memory/i.test(args),
+      // Under experimental.memoryV2, the prompt enforces an explicit
+      // one-tier-per-fact rule: team-shared project conventions (the team's
+      // test command, project-wide indentation rules) belong in the
+      // committed project-root ./GEMINI.md and must NOT be mirrored or
+      // cross-referenced into the private project memory folder
+      // (~/.gemini/tmp/<hash>/memory/). The global ~/.gemini/GEMINI.md must
+      // never be touched in this mode either.
+      await rig.waitForToolCall('write_file').catch(() => {});
+      const writeCalls = rig
+        .readToolLogs()
+        .filter((log) =>
+          ['write_file', 'replace'].includes(log.toolRequest.name),
+        );
+
+      const wroteToProjectRoot = (factPattern: RegExp) =>
+        writeCalls.some((log) => {
+          const args = log.toolRequest.args;
+          return (
+            /GEMINI\.md/i.test(args) &&
+            !/\.gemini\//i.test(args) &&
+            factPattern.test(args)
+          );
+        });
+
+      expect(
+        wroteToProjectRoot(/npm run test/i),
+        'Expected the team test-command convention to be written to the project-root ./GEMINI.md',
+      ).toBe(true);
+
+      expect(
+        wroteToProjectRoot(/2[- ]space/i),
+        'Expected the project-wide "2-space indentation" convention to be written to the project-root ./GEMINI.md',
+      ).toBe(true);
+
+      const leakedToPrivateMemory = writeCalls.some((log) => {
+        const args = log.toolRequest.args;
+        return (
+          /\.gemini\/tmp\/[^/]+\/memory\//i.test(args) &&
+          (/npm run test/i.test(args) || /2[- ]space/i.test(args))
+        );
+      });
+      expect(
+        leakedToPrivateMemory,
+        'Team-shared project conventions must NOT be mirrored into the private project memory folder (~/.gemini/tmp/<hash>/memory/) — each fact lives in exactly one tier.',
+      ).toBe(false);
+
+      const leakedToGlobal = writeCalls.some((log) => {
+        const args = log.toolRequest.args;
+        return (
+          /\.gemini\/GEMINI\.md/i.test(args) &&
+          !/tmp\/[^/]+\/memory/i.test(args)
+        );
+      });
+      expect(
+        leakedToGlobal,
+        'Project preferences must NOT be written to the global ~/.gemini/GEMINI.md',
+      ).toBe(false);
+
+      assertModelHasOutput(result);
+    },
+  });
+
+  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(
-        wasToolCalled,
-        'Expected invoke_agent to be called with save_memory agent',
+        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', {
+    suiteName: 'default',
+    suiteType: 'behavioral',
+    name: memoryV2RoutesUserProject,
+    params: {
+      settings: {
+        experimental: { memoryV2: true },
+      },
+    },
+    prompt: `Please remember my personal local dev setup for THIS project's Postgres database. This is private to my machine — do NOT commit it to the repo.
+
+Connection details:
+- Host: localhost
+- Port: 6543 (non-standard, I run multiple Postgres instances)
+- Database: myproj_dev
+- User: sandy_local
+- Password: read from the SANDY_PG_LOCAL_PASS env var in my shell
+
+How I start it locally:
+1. Run \`brew services start postgresql@15\` to bring the server up.
+2. Run \`./scripts/seed-local-db.sh\` from the repo root to load my personal seed data.
+3. Verify with \`psql -h localhost -p 6543 -U sandy_local myproj_dev -c '\\dt'\`.
+
+Quirks to remember:
+- The migrations runner sometimes hangs on my machine if I forget step 1; kill it with Ctrl+C and rerun.
+- I keep an extra \`scratch\` schema for ad-hoc experiments — never reference it from project code.`,
+    assert: async (rig, result) => {
+      // Under experimental.memoryV2 with the Private Project Memory bullet
+      // surfaced in the prompt, a fact that is project-specific AND
+      // personal-to-the-user (must not be committed) should land in the
+      // private project memory folder under ~/.gemini/tmp/<hash>/memory/. The
+      // detailed note should be written to a sibling markdown file, with
+      // MEMORY.md updated as the index. It must NOT go to committed
+      // ./GEMINI.md or the global ~/.gemini/GEMINI.md.
+      await rig.waitForToolCall('write_file').catch(() => {});
+      const writeCalls = rig
+        .readToolLogs()
+        .filter((log) =>
+          ['write_file', 'replace'].includes(log.toolRequest.name),
+        );
+
+      const wroteUserProjectDetail = writeCalls.some((log) => {
+        const args = log.toolRequest.args;
+        return (
+          /\.gemini\/tmp\/[^/]+\/memory\/(?!MEMORY\.md)[^"]+\.md/i.test(args) &&
+          /6543/.test(args)
+        );
+      });
+      expect(
+        wroteUserProjectDetail,
+        'Expected the personal-to-user project note to be written to a private project memory detail file (~/.gemini/tmp/<hash>/memory/*.md)',
+      ).toBe(true);
+
+      const wroteUserProjectIndex = writeCalls.some((log) => {
+        const args = log.toolRequest.args;
+        return /\.gemini\/tmp\/[^/]+\/memory\/MEMORY\.md/i.test(args);
+      });
+      expect(
+        wroteUserProjectIndex,
+        'Expected the personal-to-user project note to update the private project memory index (~/.gemini/tmp/<hash>/memory/MEMORY.md)',
+      ).toBe(true);
+
+      // Defensive: should NOT have written this private note to the
+      // committed project GEMINI.md or the global GEMINI.md.
+      const leakedToCommittedProject = writeCalls.some((log) => {
+        const args = log.toolRequest.args;
+        return (
+          /\/GEMINI\.md/i.test(args) &&
+          !/\.gemini\//i.test(args) &&
+          /6543/.test(args)
+        );
+      });
+      expect(
+        leakedToCommittedProject,
+        'Personal-to-user note must NOT be written to the committed project GEMINI.md',
+      ).toBe(false);
+
+      const leakedToGlobal = writeCalls.some((log) => {
+        const args = log.toolRequest.args;
+        return (
+          /\.gemini\/GEMINI\.md/i.test(args) &&
+          !/tmp\/[^/]+\/memory/i.test(args) &&
+          /6543/.test(args)
+        );
+      });
+      expect(
+        leakedToGlobal,
+        'Personal-to-user project note must NOT be written to the global ~/.gemini/GEMINI.md',
+      ).toBe(false);
+
+      assertModelHasOutput(result);
+    },
+  });
+
+  const memoryV2RoutesCrossProjectToGlobal =
+    'Agent routes cross-project personal preferences to ~/.gemini/GEMINI.md';
+  evalTest('USUALLY_PASSES', {
+    suiteName: 'default',
+    suiteType: 'behavioral',
+    name: memoryV2RoutesCrossProjectToGlobal,
+    params: {
+      settings: {
+        experimental: { memoryV2: true },
+      },
+    },
+    prompt:
+      'Please remember this about me in general: across all my projects I always prefer Prettier with single quotes and trailing commas, and I always prefer tabs over spaces for indentation. These are my personal coding-style defaults that follow me into every workspace.',
+    assert: async (rig, result) => {
+      // Under experimental.memoryV2 with the Global Personal Memory
+      // tier surfaced in the prompt, a fact that explicitly applies to the
+      // user "across all my projects" / "in every workspace" must land in
+      // the global ~/.gemini/GEMINI.md (the cross-project tier). It must
+      // NOT be mirrored into a committed project-root ./GEMINI.md (that
+      // tier is for team-shared conventions) or into the per-project
+      // private memory folder (that tier is for project-specific personal
+      // notes). Each fact lives in exactly one tier across all four tiers.
+      await rig.waitForToolCall('write_file').catch(() => {});
+      const writeCalls = rig
+        .readToolLogs()
+        .filter((log) =>
+          ['write_file', 'replace'].includes(log.toolRequest.name),
+        );
+
+      const wroteToGlobal = (factPattern: RegExp) =>
+        writeCalls.some((log) => {
+          const args = log.toolRequest.args;
+          return (
+            /\.gemini\/GEMINI\.md/i.test(args) &&
+            !/tmp\/[^/]+\/memory/i.test(args) &&
+            factPattern.test(args)
+          );
+        });
+
+      expect(
+        wroteToGlobal(/Prettier/i),
+        'Expected the cross-project Prettier preference to be written to the global personal memory file (~/.gemini/GEMINI.md)',
+      ).toBe(true);
+
+      expect(
+        wroteToGlobal(/tabs/i),
+        'Expected the cross-project "tabs over spaces" preference to be written to the global personal memory file (~/.gemini/GEMINI.md)',
+      ).toBe(true);
+
+      const leakedToCommittedProject = writeCalls.some((log) => {
+        const args = log.toolRequest.args;
+        return (
+          /GEMINI\.md/i.test(args) &&
+          !/\.gemini\//i.test(args) &&
+          (/Prettier/i.test(args) || /tabs/i.test(args))
+        );
+      });
+      expect(
+        leakedToCommittedProject,
+        'Cross-project personal preferences must NOT be mirrored into a committed project ./GEMINI.md (that tier is for team-shared conventions only)',
+      ).toBe(false);
+
+      const leakedToPrivateProject = writeCalls.some((log) => {
+        const args = log.toolRequest.args;
+        return (
+          /\.gemini\/tmp\/[^/]+\/memory\//i.test(args) &&
+          (/Prettier/i.test(args) || /tabs/i.test(args))
+        );
+      });
+      expect(
+        leakedToPrivateProject,
+        'Cross-project personal preferences must NOT be mirrored into the private project memory folder (that tier is for project-specific personal notes only)',
+      ).toBe(false);
 
       assertModelHasOutput(result);
     },

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');
@@ -126,38 +356,313 @@ 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.
- * - experimentalMemoryManager: enables the memory extraction pipeline.
+ * - experimentalAutoMemory: enables the Auto Memory skill extraction pipeline.
  * - approvalMode: YOLO auto-approves tool calls (write_file, read_file) so the
  *   background agent can execute without interactive confirmation.
  */
 const EXTRACTION_CONFIG_OVERRIDES = {
-  experimentalMemoryManager: true,
+  experimentalAutoMemory: true,
   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/subtask_delegation.eval.ts

@@ -5,10 +5,7 @@
  */
 
 import { describe, expect } from 'vitest';
-import {
-  TRACKER_CREATE_TASK_TOOL_NAME,
-  TRACKER_UPDATE_TASK_TOOL_NAME,
-} from '@google/gemini-cli-core';
+import { TRACKER_CREATE_TASK_TOOL_NAME } from '@google/gemini-cli-core';
 import { evalTest, TEST_AGENTS } from './test-helper.js';
 
 describe('subtask delegation eval test cases', () => {
@@ -22,6 +19,8 @@ describe('subtask delegation eval test cases', () => {
    * 3. Documenting (doc expert)
    */
   evalTest('USUALLY_PASSES', {
+    suiteName: 'default',
+    suiteType: 'behavioral',
     name: 'should delegate sequential subtasks to relevant experts using the task tracker',
     params: {
       settings: {
@@ -90,6 +89,8 @@ You are the doc expert. Document the provided implementation clearly.`,
    * to multiple subagents in parallel using the task tracker to manage state.
    */
   evalTest('USUALLY_PASSES', {
+    suiteName: 'default',
+    suiteType: 'behavioral',
     name: 'should delegate independent subtasks to specialists using the task tracker',
     params: {
       settings: {

evals/test-helper.ts

@@ -172,6 +172,7 @@ export async function internalEvalTest(evalCase: EvalCase) {
         timeout: evalCase.timeout,
         env: {
           GEMINI_CLI_ACTIVITY_LOG_TARGET: activityLogFile,
+          GEMINI_CLI_TRUST_WORKSPACE: 'true',
         },
       });
 

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(
@@ -119,6 +121,8 @@ describe('tracker_mode', () => {
   });
 
   evalTest('USUALLY_PASSES', {
+    suiteName: 'default',
+    suiteType: 'behavioral',
     name: 'should correctly identify the task tracker storage location from the system prompt',
     params: {
       settings: { experimental: { taskTracker: true } },
@@ -126,12 +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);
+    },
+  });
 });

evals/tsconfig.json

@@ -0,0 +1,13 @@
+{
+  "extends": "../tsconfig.json",
+  "compilerOptions": {
+    "noEmit": true,
+    "paths": {
+      "@google/gemini-cli-core": ["../packages/core/index.ts"],
+      "@google/gemini-cli": ["../packages/cli/index.ts"]
+    }
+  },
+  "include": ["**/*.ts"],
+  "exclude": ["logs"],
+  "references": [{ "path": "../packages/core" }, { "path": "../packages/cli" }]
+}

evals/unsafe-cloning.eval.ts

@@ -7,6 +7,8 @@
 import { evalTest, TestRig } from './test-helper.js';
 
 evalTest('USUALLY_PASSES', {
+  suiteName: 'default',
+  suiteType: 'behavioral',
   name: 'Reproduction: Agent uses Object.create() for cloning/delegation',
   prompt:
     'Create a utility function `createScopedConfig(config: Config, additionalDirectories: string[]): Config` in `packages/core/src/config/scoped-config.ts` that returns a new Config instance. This instance should override `getWorkspaceContext()` to include the additional directories, but delegate all other method calls (like `isPathAllowed` or `validatePathAccess`) to the original config. Note that `Config` is a complex class with private state and cannot be easily shallow-copied or reconstructed.',

evals/update_topic.eval.ts

@@ -21,6 +21,8 @@ describe('update_topic_behavior', () => {
    * more than 1/4 turns.
    */
   evalTest('USUALLY_PASSES', {
+    suiteName: 'default',
+    suiteType: 'behavioral',
     name: 'update_topic should be used at start, end and middle for complex tasks',
     prompt: `Create a simple users REST API using Express. 
 1. Initialize a new npm project and install express.
@@ -41,7 +43,7 @@ describe('update_topic_behavior', () => {
         2,
       ),
       '.gemini/settings.json': JSON.stringify({
-        experimental: {
+        general: {
           topicUpdateNarration: true,
         },
       }),
@@ -117,13 +119,15 @@ describe('update_topic_behavior', () => {
   });
 
   evalTest('USUALLY_PASSES', {
+    suiteName: 'default',
+    suiteType: 'behavioral',
     name: 'update_topic should NOT be used for informational coding tasks (Obvious)',
     approvalMode: 'default',
     prompt:
       'Explain the difference between Map and Object in JavaScript and provide a performance-focused code snippet for each.',
     files: {
       '.gemini/settings.json': JSON.stringify({
-        experimental: {
+        general: {
           topicUpdateNarration: true,
         },
       }),
@@ -142,6 +146,8 @@ describe('update_topic_behavior', () => {
   });
 
   evalTest('USUALLY_PASSES', {
+    suiteName: 'default',
+    suiteType: 'behavioral',
     name: 'update_topic should NOT be used for surgical symbol searches (Grey Area)',
     approvalMode: 'default',
     prompt:
@@ -150,7 +156,7 @@ describe('update_topic_behavior', () => {
       'packages/core/src/tools/tool-names.ts':
         "export const UPDATE_TOPIC_TOOL_NAME = 'update_topic';",
       '.gemini/settings.json': JSON.stringify({
-        experimental: {
+        general: {
           topicUpdateNarration: true,
         },
       }),
@@ -169,6 +175,8 @@ describe('update_topic_behavior', () => {
   });
 
   evalTest('USUALLY_PASSES', {
+    suiteName: 'default',
+    suiteType: 'behavioral',
     name: 'update_topic should be used for medium complexity multi-step tasks',
     prompt:
       'Refactor the `users-api` project. Move the routing logic from src/app.ts into a new file src/routes.ts, and update app.ts to use the new routes file.',
@@ -196,7 +204,7 @@ app.post('/users', (req, res) => {
 export default app;
       `,
       '.gemini/settings.json': JSON.stringify({
-        experimental: {
+        general: {
           topicUpdateNarration: true,
         },
       }),
@@ -212,7 +220,9 @@ export default app;
       expect(topicCalls.length).toBeGreaterThanOrEqual(2);
 
       // Verify it actually did the refactoring to ensure it didn't just fail immediately
-      expect(fs.existsSync(path.join(rig.testDir, 'src/routes.ts'))).toBe(true);
+      expect(fs.existsSync(path.join(rig.testDir!, 'src/routes.ts'))).toBe(
+        true,
+      );
     },
   });
 
@@ -224,6 +234,8 @@ export default app;
    * the prompt change that improves the behavior.
    */
   evalTest('USUALLY_PASSES', {
+    suiteName: 'default',
+    suiteType: 'behavioral',
     name: 'update_topic should not be called twice in a row',
     prompt: `
       We need to build a C compiler.
@@ -237,7 +249,7 @@ export default app;
     files: {
       'package.json': JSON.stringify({ name: 'test-project' }),
       '.gemini/settings.json': JSON.stringify({
-        experimental: {
+        general: {
           topicUpdateNarration: true,
         },
       }),

integration-tests/acp-telemetry.test.ts

@@ -70,6 +70,7 @@ describe('ACP telemetry', () => {
           GEMINI_API_KEY: 'fake-key',
           GEMINI_CLI_HOME: rig.homeDir!,
           GEMINI_TELEMETRY_ENABLED: 'true',
+          GEMINI_TELEMETRY_TRACES_ENABLED: 'true',
           GEMINI_TELEMETRY_TARGET: 'local',
           GEMINI_TELEMETRY_OUTFILE: telemetryPath,
         },

integration-tests/concurrency-limit.test.ts

@@ -39,7 +39,11 @@ describe('web-fetch rate limiting', () => {
     const rateLimitedCalls = toolLogs.filter(
       (log) =>
         log.toolRequest.name === 'web_fetch' &&
-        log.toolRequest.error?.includes('Rate limit exceeded'),
+        (
+          ('error' in log.toolRequest
+            ? (log.toolRequest as unknown as Record<string, string>)['error']
+            : '') as string
+        )?.includes('Rate limit exceeded'),
     );
 
     expect(rateLimitedCalls.length).toBeGreaterThan(0);

integration-tests/context-compress-interactive.test.ts

@@ -8,7 +8,16 @@ import { expect, describe, it, beforeEach, afterEach } from 'vitest';
 import { TestRig } from './test-helper.js';
 import { join } from 'node:path';
 
-describe('Interactive Mode', () => {
+// Skip on macOS: every interactive test in this file is chronically flaky
+// because the captured pty buffer contains the CLI's startup escape
+// sequences (`q4;?m...true color warning`) instead of the streamed output,
+// causing `expectText(...)` to time out. Reproducible across unrelated
+// runs on `main` (24740161950, 24739323404) and on consecutive merge-queue
+// gates for #25753 (24743605639, 24747624513) — different tests in the
+// same describe fail on different runs. Not specific to any model.
+const skipOnDarwin = process.platform === 'darwin';
+
+describe.skipIf(skipOnDarwin)('Interactive Mode', () => {
   let rig: TestRig;
 
   beforeEach(() => {

integration-tests/file-system.test.ts

@@ -134,7 +134,9 @@ describe('file-system', () => {
     ).toBeTruthy();
 
     const newFileContent = rig.readFile(fileName);
-    expect(newFileContent).toBe('hello');
+    // Trim to tolerate models that idiomatically append a trailing newline.
+    // This test is about path-with-spaces handling, not whitespace fidelity.
+    expect(newFileContent.trim()).toBe('hello');
   });
 
   it('should perform a read-then-write sequence', async () => {

integration-tests/hooks-system.test.ts

@@ -164,7 +164,8 @@ describe.skipIf(skipFlaky)(
         );
         expect(blockHook).toBeDefined();
         expect(
-          blockHook?.hookCall.stdout + blockHook?.hookCall.stderr,
+          (blockHook?.hookCall.stdout || '') +
+            (blockHook?.hookCall.stderr || ''),
         ).toContain(blockMsg);
       });
 

integration-tests/mcp-list-resources.responses

@@ -0,0 +1,2 @@
+{"method":"generateContentStream","response":[{"candidates":[{"content":{"parts":[{"functionCall":{"name":"list_mcp_resources","args":{}}}],"role":"model"},"finishReason":"STOP","index":0}],"usageMetadata":{"promptTokenCount":10,"candidatesTokenCount":10,"totalTokenCount":20}}]}
+{"method":"generateContentStream","response":[{"candidates":[{"content":{"parts":[{"text":"Here are the resources: test://resource1"}],"role":"model"},"finishReason":"STOP","index":0}],"usageMetadata":{"promptTokenCount":10,"candidatesTokenCount":10,"totalTokenCount":20}}]}

integration-tests/mcp-read-resource.responses

@@ -0,0 +1,2 @@
+{"method":"generateContentStream","response":[{"candidates":[{"content":{"parts":[{"functionCall":{"name":"read_mcp_resource","args":{"uri":"test://resource1"}}}],"role":"model"},"finishReason":"STOP","index":0}],"usageMetadata":{"promptTokenCount":10,"candidatesTokenCount":10,"totalTokenCount":20}}]}
+{"method":"generateContentStream","response":[{"candidates":[{"content":{"parts":[{"text":"The content is: content of resource 1"}],"role":"model"},"finishReason":"STOP","index":0}],"usageMetadata":{"promptTokenCount":10,"candidatesTokenCount":10,"totalTokenCount":20}}]}

integration-tests/mcp-resources.responses

@@ -0,0 +1,4 @@
+{"method":"generateContentStream","response":[{"candidates":[{"content":{"parts":[{"functionCall":{"name":"list_mcp_resources","args":{}}}],"role":"model"},"finishReason":"STOP","index":0}],"usageMetadata":{"promptTokenCount":10,"candidatesTokenCount":10,"totalTokenCount":20}}]}
+{"method":"generateContentStream","response":[{"candidates":[{"content":{"parts":[{"text":"Here are the resources: test://resource1"}],"role":"model"},"finishReason":"STOP","index":0}],"usageMetadata":{"promptTokenCount":10,"candidatesTokenCount":10,"totalTokenCount":20}}]}
+{"method":"generateContentStream","response":[{"candidates":[{"content":{"parts":[{"functionCall":{"name":"read_mcp_resource","args":{"uri":"test://resource1"}}}],"role":"model"},"finishReason":"STOP","index":0}],"usageMetadata":{"promptTokenCount":10,"candidatesTokenCount":10,"totalTokenCount":20}}]}
+{"method":"generateContentStream","response":[{"candidates":[{"content":{"parts":[{"text":"The content is: content of resource 1"}],"role":"model"},"finishReason":"STOP","index":0}],"usageMetadata":{"promptTokenCount":10,"candidatesTokenCount":10,"totalTokenCount":20}}]}

integration-tests/mcp-resources.test.ts

@@ -0,0 +1,178 @@
+/**
+ * @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 { join, dirname } from 'node:path';
+import { fileURLToPath } from 'node:url';
+import fs from 'node:fs';
+
+const __dirname = dirname(fileURLToPath(import.meta.url));
+
+describe('mcp-resources-integration', () => {
+  let rig: TestRig;
+
+  beforeEach(() => {
+    rig = new TestRig();
+  });
+
+  afterEach(async () => await rig.cleanup());
+
+  it('should list mcp resources', async () => {
+    await rig.setup('mcp-list-resources-test', {
+      settings: {
+        model: {
+          name: 'gemini-3-flash-preview',
+        },
+      },
+      fakeResponsesPath: join(__dirname, 'mcp-list-resources.responses'),
+    });
+
+    // Workaround for ProjectRegistry save issue
+    const userGeminiDir = join(rig.homeDir!, '.gemini');
+    fs.writeFileSync(join(userGeminiDir, 'projects.json'), '{"projects":{}}');
+
+    // Add a dummy server to get setup done
+    rig.addTestMcpServer('resource-server', {
+      name: 'resource-server',
+      tools: [],
+    });
+
+    // Overwrite the script with resource support
+    const scriptPath = join(rig.testDir!, 'test-mcp-resource-server.mjs');
+    const scriptContent = `
+import { Server } from '@modelcontextprotocol/sdk/server/index.js';
+import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js';
+import {
+  ListResourcesRequestSchema,
+} from '@modelcontextprotocol/sdk/types.js';
+
+const server = new Server(
+  {
+    name: 'resource-server',
+    version: '1.0.0',
+  },
+  {
+    capabilities: {
+      resources: {},
+    },
+  },
+);
+
+server.setRequestHandler(ListResourcesRequestSchema, async () => {
+  return {
+    resources: [
+      {
+        uri: 'test://resource1',
+        name: 'Resource 1',
+        mimeType: 'text/plain',
+        description: 'A test resource',
+      }
+    ],
+  };
+});
+
+const transport = new StdioServerTransport();
+await server.connect(transport);
+`;
+    fs.writeFileSync(scriptPath, scriptContent);
+
+    const output = await rig.run({
+      args: 'List all available MCP resources.',
+      env: { GEMINI_API_KEY: 'dummy' },
+    });
+
+    const foundCall = await rig.waitForToolCall('list_mcp_resources');
+    expect(foundCall).toBeTruthy();
+    expect(output).toContain('test://resource1');
+  }, 60000);
+
+  it('should read mcp resource', async () => {
+    await rig.setup('mcp-read-resource-test', {
+      settings: {
+        model: {
+          name: 'gemini-3-flash-preview',
+        },
+      },
+      fakeResponsesPath: join(__dirname, 'mcp-read-resource.responses'),
+    });
+
+    // Workaround for ProjectRegistry save issue
+    const userGeminiDir = join(rig.homeDir!, '.gemini');
+    fs.writeFileSync(join(userGeminiDir, 'projects.json'), '{"projects":{}}');
+
+    // Add a dummy server to get setup done
+    rig.addTestMcpServer('resource-server', {
+      name: 'resource-server',
+      tools: [],
+    });
+
+    // Overwrite the script with resource support
+    const scriptPath = join(rig.testDir!, 'test-mcp-resource-server.mjs');
+    const scriptContent = `
+import { Server } from '@modelcontextprotocol/sdk/server/index.js';
+import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js';
+import {
+  ListResourcesRequestSchema,
+  ReadResourceRequestSchema,
+} from '@modelcontextprotocol/sdk/types.js';
+
+const server = new Server(
+  {
+    name: 'resource-server',
+    version: '1.0.0',
+  },
+  {
+    capabilities: {
+      resources: {},
+    },
+  },
+);
+
+// Need to provide list resources so the tool is active!
+server.setRequestHandler(ListResourcesRequestSchema, async () => {
+  return {
+    resources: [
+      {
+        uri: 'test://resource1',
+        name: 'Resource 1',
+        mimeType: 'text/plain',
+        description: 'A test resource',
+      }
+    ],
+  };
+});
+
+server.setRequestHandler(ReadResourceRequestSchema, async (request) => {
+  if (request.params.uri === 'test://resource1') {
+    return {
+      contents: [
+        {
+          uri: 'test://resource1',
+          mimeType: 'text/plain',
+          text: 'This is the content of resource 1',
+        }
+      ],
+    };
+  }
+  throw new Error('Resource not found');
+});
+
+const transport = new StdioServerTransport();
+await server.connect(transport);
+`;
+    fs.writeFileSync(scriptPath, scriptContent);
+
+    const output = await rig.run({
+      args: 'Read the MCP resource test://resource1.',
+      env: { GEMINI_API_KEY: 'dummy' },
+    });
+
+    const foundCall = await rig.waitForToolCall('read_mcp_resource');
+    expect(foundCall).toBeTruthy();
+    expect(output).toContain('content of resource 1');
+  }, 60000);
+});

integration-tests/plan-mode.test.ts

@@ -81,7 +81,10 @@ describe('Plan Mode', () => {
 
     await rig.run({
       approvalMode: 'plan',
-      args: 'Create a file called plan.md in the plans directory.',
+      args:
+        'Create a file called plan.md in the plans directory with the ' +
+        'content "# Plan". Treat this as a Directive and write the file ' +
+        'immediately without proposing strategy or asking for confirmation.',
     });
 
     const toolLogs = rig.readToolLogs();
@@ -108,7 +111,7 @@ describe('Plan Mode', () => {
     ).toBeDefined();
     expect(
       planWrite?.toolRequest.success,
-      `Expected write_file to succeed, but it failed with error: ${planWrite?.toolRequest.error}`,
+      `Expected write_file to succeed, but it failed with error: ${'error' in (planWrite?.toolRequest || {}) ? (planWrite?.toolRequest as unknown as Record<string, string>)['error'] : 'unknown'}`,
     ).toBe(true);
   });
 
@@ -194,7 +197,11 @@ describe('Plan Mode', () => {
 
     await rig.run({
       approvalMode: 'plan',
-      args: 'Create a file called plan-no-session.md in the plans directory.',
+      args:
+        'Create a file called plan-no-session.md in the plans directory ' +
+        'with the content "# Plan". Treat this as a Directive and write ' +
+        'the file immediately without proposing strategy or asking for ' +
+        'confirmation.',
     });
 
     const toolLogs = rig.readToolLogs();
@@ -221,7 +228,7 @@ describe('Plan Mode', () => {
     ).toBeDefined();
     expect(
       planWrite?.toolRequest.success,
-      `Expected write_file to succeed, but it failed with error: ${planWrite?.toolRequest.error}`,
+      `Expected write_file to succeed, but it failed with error: ${'error' in (planWrite?.toolRequest || {}) ? (planWrite?.toolRequest as unknown as Record<string, string>)['error'] : 'unknown'}`,
     ).toBe(true);
   });
   it('should switch from a pro model to a flash model after exiting plan mode', async () => {
@@ -270,13 +277,24 @@ describe('Plan Mode', () => {
     );
 
     const apiRequests = rig.readAllApiRequest();
-    const modelNames = apiRequests.map((r) => r.attributes?.model || 'unknown');
+    const modelNames = apiRequests.map(
+      (r) =>
+        ('model' in (r.attributes || {})
+          ? (r.attributes as unknown as Record<string, string>)['model']
+          : 'unknown') || 'unknown',
+    );
 
     const proRequests = apiRequests.filter((r) =>
-      r.attributes?.model?.includes('pro'),
+      ('model' in (r.attributes || {})
+        ? (r.attributes as unknown as Record<string, string>)['model']
+        : 'unknown'
+      )?.includes('pro'),
     );
     const flashRequests = apiRequests.filter((r) =>
-      r.attributes?.model?.includes('flash'),
+      ('model' in (r.attributes || {})
+        ? (r.attributes as unknown as Record<string, string>)['model']
+        : 'unknown'
+      )?.includes('flash'),
     );
 
     expect(

integration-tests/tsconfig.json

@@ -5,5 +5,9 @@
     "allowJs": true
   },
   "include": ["**/*.ts"],
-  "references": [{ "path": "../packages/core" }]
+  "references": [
+    { "path": "../packages/core" },
+    { "path": "../packages/test-utils" },
+    { "path": "../packages/cli" }
+  ]
 }

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();
+  });
+});

memory-tests/baselines.json

@@ -1,55 +1,55 @@
 {
   "version": 1,
-  "updatedAt": "2026-04-10T15:36:04.547Z",
+  "updatedAt": "2026-04-20T18:04:59.671Z",
   "scenarios": {
     "multi-turn-conversation": {
-      "heapUsedBytes": 120082704,
-      "heapTotalBytes": 177586176,
-      "rssBytes": 269172736,
-      "externalBytes": 4304053,
-      "timestamp": "2026-04-10T15:35:17.603Z"
+      "heapUsedMB": 68.8,
+      "heapTotalMB": 91.2,
+      "rssMB": 215.4,
+      "externalMB": 93.8,
+      "timestamp": "2026-04-20T18:02:40.101Z"
     },
     "multi-function-call-repo-search": {
-      "heapUsedBytes": 104644984,
-      "heapTotalBytes": 111575040,
-      "rssBytes": 204079104,
-      "externalBytes": 4304053,
-      "timestamp": "2026-04-10T15:35:22.480Z"
+      "heapUsedMB": 73.5,
+      "heapTotalMB": 93.1,
+      "rssMB": 223.6,
+      "externalMB": 97.7,
+      "timestamp": "2026-04-20T18:02:42.032Z"
     },
     "idle-session-startup": {
-      "heapUsedBytes": 119813672,
-      "heapTotalBytes": 177061888,
-      "rssBytes": 267943936,
-      "externalBytes": 4304053,
-      "timestamp": "2026-04-10T15:35:08.035Z"
+      "heapUsedMB": 69.8,
+      "heapTotalMB": 92.4,
+      "rssMB": 217.4,
+      "externalMB": 93.8,
+      "timestamp": "2026-04-20T18:02:36.294Z"
     },
     "simple-prompt-response": {
-      "heapUsedBytes": 119722064,
-      "heapTotalBytes": 177324032,
-      "rssBytes": 268812288,
-      "externalBytes": 4304053,
-      "timestamp": "2026-04-10T15:35:12.770Z"
+      "heapUsedMB": 69.5,
+      "heapTotalMB": 92.4,
+      "rssMB": 216.1,
+      "externalMB": 93.8,
+      "timestamp": "2026-04-20T18:02:38.198Z"
     },
     "resume-large-chat-with-messages": {
-      "heapUsedBytes": 106545568,
-      "heapTotalBytes": 111509504,
-      "rssBytes": 202596352,
-      "externalBytes": 4306101,
-      "timestamp": "2026-04-10T15:36:04.547Z"
+      "heapUsedMB": 887.1,
+      "heapTotalMB": 954.3,
+      "rssMB": 1109.6,
+      "externalMB": 103.2,
+      "timestamp": "2026-04-20T18:04:59.671Z"
     },
     "resume-large-chat": {
-      "heapUsedBytes": 106513760,
-      "heapTotalBytes": 111509504,
-      "rssBytes": 202596352,
-      "externalBytes": 4306101,
-      "timestamp": "2026-04-10T15:35:59.528Z"
+      "heapUsedMB": 885.6,
+      "heapTotalMB": 955.6,
+      "rssMB": 1107.8,
+      "externalMB": 110.5,
+      "timestamp": "2026-04-20T18:04:06.526Z"
     },
     "large-chat": {
-      "heapUsedBytes": 106471568,
-      "heapTotalBytes": 111509504,
-      "rssBytes": 202596352,
-      "externalBytes": 4306101,
-      "timestamp": "2026-04-10T15:35:53.180Z"
+      "heapUsedMB": 158.5,
+      "heapTotalMB": 193,
+      "rssMB": 787.9,
+      "externalMB": 104,
+      "timestamp": "2026-04-20T18:03:12.486Z"
     }
   }
 }

memory-tests/memory-usage.test.ts

@@ -16,15 +16,21 @@ import {
   mkdirSync,
   rmSync,
 } from 'node:fs';
-import { randomUUID } from 'node:crypto';
+import { randomUUID, createHash } from 'node:crypto';
 
 const __dirname = dirname(fileURLToPath(import.meta.url));
 const BASELINES_PATH = join(__dirname, 'baselines.json');
 const UPDATE_BASELINES = process.env['UPDATE_MEMORY_BASELINES'] === 'true';
+function getProjectHash(projectRoot: string): string {
+  return createHash('sha256').update(projectRoot).digest('hex');
+}
 const TOLERANCE_PERCENT = 10;
 
 // Fake API key for tests using fake responses
-const TEST_ENV = { GEMINI_API_KEY: 'fake-memory-test-key' };
+const TEST_ENV = {
+  GEMINI_API_KEY: 'fake-memory-test-key',
+  GEMINI_MEMORY_MONITOR_INTERVAL: '100',
+};
 
 describe('Memory Usage Tests', () => {
   let harness: MemoryTestHarness;
@@ -56,6 +62,7 @@ describe('Memory Usage Tests', () => {
     });
 
     const result = await harness.runScenario(
+      rig,
       'idle-session-startup',
       async (recordSnapshot) => {
         await rig.run({
@@ -85,6 +92,7 @@ describe('Memory Usage Tests', () => {
     });
 
     const result = await harness.runScenario(
+      rig,
       'simple-prompt-response',
       async (recordSnapshot) => {
         await rig.run({
@@ -122,6 +130,7 @@ describe('Memory Usage Tests', () => {
     ];
 
     const result = await harness.runScenario(
+      rig,
       'multi-turn-conversation',
       async (recordSnapshot) => {
         // Run through all turns as a piped sequence
@@ -144,6 +153,9 @@ describe('Memory Usage Tests', () => {
       );
     } else {
       harness.assertWithinBaseline(result);
+      harness.assertMemoryReturnsToBaseline(result.snapshots, 20);
+      const { leaked, message } = harness.analyzeSnapshots(result.snapshots);
+      if (leaked) console.warn(`⚠ ${message}`);
     }
   });
 
@@ -168,6 +180,7 @@ describe('Memory Usage Tests', () => {
     );
 
     const result = await harness.runScenario(
+      rig,
       'multi-function-call-repo-search',
       async (recordSnapshot) => {
         await rig.run({
@@ -189,6 +202,7 @@ describe('Memory Usage Tests', () => {
       );
     } else {
       harness.assertWithinBaseline(result);
+      harness.assertMemoryReturnsToBaseline(result.snapshots, 20);
     }
   });
 
@@ -228,6 +242,7 @@ describe('Memory Usage Tests', () => {
       });
 
       const result = await harness.runScenario(
+        rig,
         'large-chat',
         async (recordSnapshot) => {
           await rig.run({
@@ -257,19 +272,21 @@ describe('Memory Usage Tests', () => {
       });
 
       const result = await harness.runScenario(
+        rig,
         'resume-large-chat',
         async (recordSnapshot) => {
           // Ensure the history file is linked
           const targetChatsDir = join(
-            rig.testDir!,
+            rig.homeDir!,
+            '.gemini',
             'tmp',
-            'test-project-hash',
+            getProjectHash(rig.testDir!),
             'chats',
           );
           mkdirSync(targetChatsDir, { recursive: true });
           const targetHistoryPath = join(
             targetChatsDir,
-            'large-chat-session.json',
+            'session-large-chat.json',
           );
           if (existsSync(targetHistoryPath)) rmSync(targetHistoryPath);
           copyFileSync(sharedHistoryPath, targetHistoryPath);
@@ -302,19 +319,21 @@ describe('Memory Usage Tests', () => {
       });
 
       const result = await harness.runScenario(
+        rig,
         'resume-large-chat-with-messages',
         async (recordSnapshot) => {
           // Ensure the history file is linked
           const targetChatsDir = join(
-            rig.testDir!,
+            rig.homeDir!,
+            '.gemini',
             'tmp',
-            'test-project-hash',
+            getProjectHash(rig.testDir!),
             'chats',
           );
           mkdirSync(targetChatsDir, { recursive: true });
           const targetHistoryPath = join(
             targetChatsDir,
-            'large-chat-session.json',
+            'session-large-chat.json',
           );
           if (existsSync(targetHistoryPath)) rmSync(targetHistoryPath);
           copyFileSync(sharedHistoryPath, targetHistoryPath);
@@ -457,6 +476,9 @@ async function generateSharedLargeChatData(tempDir: string) {
   // Generate responses for resumed chat
   const resumeResponsesStream = createWriteStream(resumeResponsesPath);
   for (let i = 0; i < 5; i++) {
+    // Doubling up on non-streaming responses to satisfy classifier and complexity checks
+    resumeResponsesStream.write(JSON.stringify(complexityResponse) + '\n');
+    resumeResponsesStream.write(JSON.stringify(summaryResponse) + '\n');
     resumeResponsesStream.write(JSON.stringify(complexityResponse) + '\n');
     resumeResponsesStream.write(
       JSON.stringify({
@@ -489,8 +511,12 @@ async function generateSharedLargeChatData(tempDir: string) {
 
   // Wait for streams to finish
   await Promise.all([
-    new Promise((res) => activeResponsesStream.on('finish', res)),
-    new Promise((res) => resumeResponsesStream.on('finish', res)),
+    new Promise((res) =>
+      activeResponsesStream.on('finish', () => res(undefined)),
+    ),
+    new Promise((res) =>
+      resumeResponsesStream.on('finish', () => res(undefined)),
+    ),
   ]);
 
   return {

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@google/gemini-cli",
-  "version": "0.40.0-nightly.20260414.g5b1f7375a",
+  "version": "0.41.0-nightly.20260423.gaa05b4583",
   "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:0.40.0-nightly.20260414.g5b1f7375a"
+    "sandboxImageUri": "us-docker.pkg.dev/gemini-code-dev/gemini-cli/sandbox:0.41.0-nightly.20260423.gaa05b4583"
   },
   "scripts": {
     "start": "cross-env NODE_ENV=development node scripts/start.js",
@@ -62,7 +62,7 @@
     "lint:ci": "npm run lint:all",
     "lint:all": "node scripts/lint.js",
     "format": "prettier --experimental-cli --write .",
-    "typecheck": "npm run typecheck --workspaces --if-present",
+    "typecheck": "npm run typecheck --workspaces --if-present && tsc -b evals/tsconfig.json integration-tests/tsconfig.json memory-tests/tsconfig.json",
     "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",
@@ -81,8 +81,7 @@
     "glob": "^12.0.0",
     "node-domexception": "npm:empty@^0.10.1",
     "prebuild-install": "npm:nop@1.0.0",
-    "cross-spawn": "^7.0.6",
-    "minimatch": "^10.2.2"
+    "cross-spawn": "^7.0.6"
   },
   "bin": {
     "gemini": "bundle/gemini.js"
@@ -94,6 +93,7 @@
   ],
   "devDependencies": {
     "@agentclientprotocol/sdk": "^0.16.1",
+    "read-package-up": "^11.0.0",
     "@octokit/rest": "^22.0.0",
     "@types/marked": "^5.0.2",
     "@types/mime-types": "^3.0.1",
@@ -137,6 +137,7 @@
     "strip-ansi": "^7.1.2",
     "ts-prune": "^0.10.3",
     "tsx": "^4.20.3",
+    "typescript": "^5.8.3",
     "typescript-eslint": "^8.30.1",
     "vitest": "^3.2.4",
     "yargs": "^17.7.2"

packages/a2a-server/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@google/gemini-cli-a2a-server",
-  "version": "0.40.0-nightly.20260414.g5b1f7375a",
+  "version": "0.41.0-nightly.20260423.gaa05b4583",
   "description": "Gemini CLI A2A Server",
   "repository": {
     "type": "git",

packages/a2a-server/src/utils/testing_utils.ts

@@ -98,6 +98,7 @@ export function createMockConfig(
       getMcpServers: vi.fn().mockReturnValue({}),
     }),
     getTelemetryLogPromptsEnabled: vi.fn().mockReturnValue(false),
+    getTelemetryTracesEnabled: vi.fn().mockReturnValue(false),
     getGitService: vi.fn(),
     validatePathAccess: vi.fn().mockReturnValue(undefined),
     getShellExecutionConfig: vi.fn().mockReturnValue({
@@ -111,6 +112,7 @@ export function createMockConfig(
     }),
     isContextManagementEnabled: vi.fn().mockReturnValue(false),
     getContextManagementConfig: vi.fn().mockReturnValue({ enabled: false }),
+    getExperimentalGemma: vi.fn().mockReturnValue(false),
     ...overrides,
   } as unknown as Config;
 

packages/cli/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@google/gemini-cli",
-  "version": "0.40.0-nightly.20260414.g5b1f7375a",
+  "version": "0.41.0-nightly.20260423.gaa05b4583",
   "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:0.40.0-nightly.20260414.g5b1f7375a"
+    "sandboxImageUri": "us-docker.pkg.dev/gemini-code-dev/gemini-cli/sandbox:0.41.0-nightly.20260423.gaa05b4583"
   },
   "dependencies": {
     "@agentclientprotocol/sdk": "^0.16.1",

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

@@ -41,6 +41,8 @@ import * as fs from 'node:fs/promises';
 import * as path from 'node:path';
 import { ApprovalMode } from '@google/gemini-cli-core/src/policy/types.js';
 
+const startMemoryServiceMock = vi.hoisted(() => vi.fn());
+
 vi.mock('../config/config.js', () => ({
   loadCliConfig: vi.fn(),
 }));
@@ -101,6 +103,7 @@ vi.mock(
     const actual = await importOriginal();
     return {
       ...actual,
+      startMemoryService: startMemoryServiceMock,
       updatePolicy: vi.fn(),
       createPolicyUpdater: vi.fn(),
       ReadManyFilesTool: vi.fn(),
@@ -148,6 +151,8 @@ describe('GeminiAgent', () => {
   let agent: GeminiAgent;
 
   beforeEach(() => {
+    vi.clearAllMocks();
+    startMemoryServiceMock.mockResolvedValue(undefined);
     mockConfig = {
       refreshAuth: vi.fn(),
       initialize: vi.fn(),
@@ -155,6 +160,7 @@ describe('GeminiAgent', () => {
       getFileSystemService: vi.fn(),
       setFileSystemService: vi.fn(),
       getContentGeneratorConfig: vi.fn(),
+      isAutoMemoryEnabled: vi.fn().mockReturnValue(false),
       getActiveModel: vi.fn().mockReturnValue('gemini-pro'),
       getModel: vi.fn().mockReturnValue('gemini-pro'),
       getGeminiClient: vi.fn().mockReturnValue({
@@ -354,6 +360,34 @@ describe('GeminiAgent', () => {
     vi.useRealTimers();
   });
 
+  it('should start auto memory for new ACP sessions when enabled', async () => {
+    mockConfig.getContentGeneratorConfig = vi.fn().mockReturnValue({
+      apiKey: 'test-key',
+    });
+    mockConfig.isAutoMemoryEnabled = vi.fn().mockReturnValue(true);
+
+    await agent.newSession({
+      cwd: '/tmp',
+      mcpServers: [],
+    });
+
+    expect(startMemoryServiceMock).toHaveBeenCalledWith(mockConfig);
+  });
+
+  it('should not start auto memory for new ACP sessions when disabled', async () => {
+    mockConfig.getContentGeneratorConfig = vi.fn().mockReturnValue({
+      apiKey: 'test-key',
+    });
+    mockConfig.isAutoMemoryEnabled = vi.fn().mockReturnValue(false);
+
+    await agent.newSession({
+      cwd: '/tmp',
+      mcpServers: [],
+    });
+
+    expect(startMemoryServiceMock).not.toHaveBeenCalled();
+  });
+
   it('should return modes without plan mode when plan is disabled', async () => {
     mockConfig.getContentGeneratorConfig = vi.fn().mockReturnValue({
       apiKey: 'test-key',

packages/cli/src/acp/acpClient.ts

@@ -76,6 +76,7 @@ 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';
 
@@ -324,6 +325,7 @@ export class GeminiAgent {
 
     await config.initialize();
     startupProfiler.flush(config);
+    startAutoMemoryIfEnabled(config);
 
     const geminiClient = config.getGeminiClient();
     const chat = await geminiClient.startChat();
@@ -465,6 +467,7 @@ export class GeminiAgent {
     // which starts the MCP servers and other heavy resources.
     await config.initialize();
     startupProfiler.flush(config);
+    startAutoMemoryIfEnabled(config);
 
     return config;
   }

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

@@ -100,6 +100,7 @@ describe('GeminiAgent Session Resume', () => {
         unsubscribe: vi.fn(),
       },
       getApprovalMode: vi.fn().mockReturnValue('default'),
+      isAutoMemoryEnabled: vi.fn().mockReturnValue(false),
       isPlanEnabled: vi.fn().mockReturnValue(true),
       getModel: vi.fn().mockReturnValue('gemini-pro'),
       getHasAccessToPreviewModel: vi.fn().mockReturnValue(false),

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

@@ -135,10 +135,10 @@ export class InboxMemoryCommand implements Command {
     context: CommandContext,
     _: string[],
   ): Promise<CommandExecutionResponse> {
-    if (!context.agentContext.config.isMemoryManagerEnabled()) {
+    if (!context.agentContext.config.isAutoMemoryEnabled()) {
       return {
         name: this.name,
-        data: 'The memory inbox requires the experimental memory manager. Enable it with: experimental.memoryManager = true in settings.',
+        data: 'The memory inbox requires Auto Memory. Enable it with: experimental.autoMemory = true in settings.',
       };
     }
 

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

@@ -0,0 +1,247 @@
+/**
+ * @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;
+
+  beforeEach(() => {
+    vi.resetAllMocks();
+    listCommand = new ListCheckpointsCommand();
+    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 () => {
+    vi.mocked(fs.readdir).mockResolvedValue([
+      'not-a-checkpoint.txt',
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    ] as any);
+
+    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'));
+    vi.mocked(fs.readdir).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 () => {
+    vi.mocked(fs.readdir).mockResolvedValue([
+      'cp1.json',
+      'cp2.json',
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    ] as any);
+    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 () => {
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    vi.mocked(fs.readdir).mockResolvedValue(['some.json'] as any);
+    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 () => {
+    vi.mocked(fs.readdir).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/commands/gemma.ts

@@ -0,0 +1,33 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { CommandModule, Argv } from 'yargs';
+import { initializeOutputListenersAndFlush } from '../gemini.js';
+import { defer } from '../deferred.js';
+import { setupCommand } from './gemma/setup.js';
+import { startCommand } from './gemma/start.js';
+import { stopCommand } from './gemma/stop.js';
+import { statusCommand } from './gemma/status.js';
+import { logsCommand } from './gemma/logs.js';
+
+export const gemmaCommand: CommandModule = {
+  command: 'gemma',
+  describe: 'Manage local Gemma model routing',
+  builder: (yargs: Argv) =>
+    yargs
+      .middleware((argv) => {
+        initializeOutputListenersAndFlush();
+        argv['isCommand'] = true;
+      })
+      .command(defer(setupCommand, 'gemma'))
+      .command(defer(startCommand, 'gemma'))
+      .command(defer(stopCommand, 'gemma'))
+      .command(defer(statusCommand, 'gemma'))
+      .command(defer(logsCommand, 'gemma'))
+      .demandCommand(1, 'You need at least one command before continuing.')
+      .version(false),
+  handler: () => {},
+};

packages/cli/src/commands/gemma/constants.ts

@@ -0,0 +1,45 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import path from 'node:path';
+import { Storage } from '@google/gemini-cli-core';
+
+export const LITERT_RELEASE_VERSION = 'v0.9.0-alpha03';
+export const LITERT_RELEASE_BASE_URL =
+  'https://github.com/google-ai-edge/LiteRT-LM/releases/download';
+export const GEMMA_MODEL_NAME = 'gemma3-1b-gpu-custom';
+export const DEFAULT_PORT = 9379;
+export const HEALTH_CHECK_TIMEOUT_MS = 5000;
+export const LITERT_API_VERSION = 'v1beta';
+export const SERVER_START_WAIT_MS = 3000;
+
+export const PLATFORM_BINARY_MAP: Record<string, string> = {
+  'darwin-arm64': 'lit.macos_arm64',
+  'linux-x64': 'lit.linux_x86_64',
+  'win32-x64': 'lit.windows_x86_64.exe',
+};
+
+// SHA-256 hashes for the official LiteRT-LM v0.9.0-alpha03 release binaries.
+export const PLATFORM_BINARY_SHA256: Record<string, string> = {
+  'lit.macos_arm64':
+    '9e826a2634f2e8b220ad0f1e1b5c139e0b47cb172326e3b7d46d31382f49478e',
+  'lit.linux_x86_64':
+    '66601df8a07f08244b188e9fcab0bf4a16562fe76d8d47e49f40273d57541ee8',
+  'lit.windows_x86_64.exe':
+    'de82d2829d2fb1cbdb318e2d8a78dc2f9659ff14cb11b2894d1f30e0bfde2bf6',
+};
+
+export function getLiteRtBinDir(): string {
+  return path.join(Storage.getGlobalGeminiDir(), 'bin', 'litert');
+}
+
+export function getPidFilePath(): string {
+  return path.join(Storage.getGlobalTempDir(), 'litert-server.pid');
+}
+
+export function getLogFilePath(): string {
+  return path.join(Storage.getGlobalTempDir(), 'litert-server.log');
+}

packages/cli/src/commands/gemma/logs.test.ts

@@ -0,0 +1,186 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import fs from 'node:fs';
+import type { ChildProcess } from 'node:child_process';
+import { EventEmitter } from 'node:events';
+import os from 'node:os';
+import path from 'node:path';
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
+import { spawn } from 'node:child_process';
+import { exitCli } from '../utils.js';
+import { getLogFilePath } from './constants.js';
+import { logsCommand, readLastLines } from './logs.js';
+
+vi.mock('@google/gemini-cli-core', async (importOriginal) => {
+  const { mockCoreDebugLogger } = await import(
+    '../../test-utils/mockDebugLogger.js'
+  );
+  return mockCoreDebugLogger(
+    await importOriginal<typeof import('@google/gemini-cli-core')>(),
+    {
+      stripAnsi: false,
+    },
+  );
+});
+
+vi.mock('node:child_process', async (importOriginal) => {
+  const actual = await importOriginal<typeof import('node:child_process')>();
+  return {
+    ...actual,
+    spawn: vi.fn(),
+  };
+});
+
+vi.mock('../utils.js', () => ({
+  exitCli: vi.fn(),
+}));
+
+vi.mock('./constants.js', () => ({
+  getLogFilePath: vi.fn(),
+}));
+
+function createMockChild(): ChildProcess {
+  return Object.assign(new EventEmitter(), {
+    kill: vi.fn(),
+  }) as unknown as ChildProcess;
+}
+
+async function flushMicrotasks() {
+  await Promise.resolve();
+  await Promise.resolve();
+}
+
+describe('readLastLines', () => {
+  const tempFiles: string[] = [];
+
+  afterEach(async () => {
+    await Promise.all(
+      tempFiles
+        .splice(0)
+        .map((filePath) => fs.promises.rm(filePath, { force: true })),
+    );
+  });
+
+  it('returns only the requested tail lines without reading the whole file eagerly', async () => {
+    const filePath = path.join(
+      os.tmpdir(),
+      `gemma-logs-${Date.now()}-${Math.random().toString(36).slice(2)}.log`,
+    );
+    tempFiles.push(filePath);
+
+    const content = Array.from({ length: 2000 }, (_, i) => `line-${i + 1}`)
+      .join('\n')
+      .concat('\n');
+    await fs.promises.writeFile(filePath, content, 'utf-8');
+
+    await expect(readLastLines(filePath, 3)).resolves.toBe(
+      'line-1998\nline-1999\nline-2000\n',
+    );
+  });
+
+  it('returns an empty string when zero lines are requested', async () => {
+    const filePath = path.join(
+      os.tmpdir(),
+      `gemma-logs-${Date.now()}-${Math.random().toString(36).slice(2)}.log`,
+    );
+    tempFiles.push(filePath);
+    await fs.promises.writeFile(filePath, 'line-1\nline-2\n', 'utf-8');
+
+    await expect(readLastLines(filePath, 0)).resolves.toBe('');
+  });
+});
+
+describe('logsCommand', () => {
+  const originalPlatform = process.platform;
+
+  beforeEach(() => {
+    vi.clearAllMocks();
+    Object.defineProperty(process, 'platform', {
+      value: 'linux',
+      configurable: true,
+    });
+    vi.mocked(getLogFilePath).mockReturnValue('/tmp/gemma.log');
+    vi.spyOn(fs.promises, 'access').mockResolvedValue(undefined);
+  });
+
+  afterEach(() => {
+    Object.defineProperty(process, 'platform', {
+      value: originalPlatform,
+      configurable: true,
+    });
+    vi.restoreAllMocks();
+  });
+
+  it('waits for the tail process to close before exiting in follow mode', async () => {
+    const child = createMockChild();
+    vi.mocked(spawn).mockReturnValue(child);
+
+    let resolved = false;
+    const handlerPromise = (
+      logsCommand.handler as (argv: Record<string, unknown>) => Promise<void>
+    )({}).then(() => {
+      resolved = true;
+    });
+
+    await flushMicrotasks();
+
+    expect(spawn).toHaveBeenCalledWith(
+      'tail',
+      ['-f', '-n', '20', '/tmp/gemma.log'],
+      { stdio: 'inherit' },
+    );
+    expect(resolved).toBe(false);
+    expect(exitCli).not.toHaveBeenCalled();
+
+    child.emit('close', 0);
+    await handlerPromise;
+
+    expect(exitCli).toHaveBeenCalledWith(0);
+  });
+
+  it('uses one-shot tail output when follow is disabled', async () => {
+    const child = createMockChild();
+    vi.mocked(spawn).mockReturnValue(child);
+
+    const handlerPromise = (
+      logsCommand.handler as (argv: Record<string, unknown>) => Promise<void>
+    )({ follow: false });
+
+    await flushMicrotasks();
+
+    expect(spawn).toHaveBeenCalledWith('tail', ['-n', '20', '/tmp/gemma.log'], {
+      stdio: 'inherit',
+    });
+
+    child.emit('close', 0);
+    await handlerPromise;
+
+    expect(exitCli).toHaveBeenCalledWith(0);
+  });
+
+  it('follows from the requested line count when both --lines and --follow are set', async () => {
+    const child = createMockChild();
+    vi.mocked(spawn).mockReturnValue(child);
+
+    const handlerPromise = (
+      logsCommand.handler as (argv: Record<string, unknown>) => Promise<void>
+    )({ lines: 5, follow: true });
+
+    await flushMicrotasks();
+
+    expect(spawn).toHaveBeenCalledWith(
+      'tail',
+      ['-f', '-n', '5', '/tmp/gemma.log'],
+      { stdio: 'inherit' },
+    );
+
+    child.emit('close', 0);
+    await handlerPromise;
+
+    expect(exitCli).toHaveBeenCalledWith(0);
+  });
+});

packages/cli/src/commands/gemma/logs.ts

@@ -0,0 +1,200 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { CommandModule } from 'yargs';
+import fs from 'node:fs';
+import { spawn, type ChildProcess } from 'node:child_process';
+import { debugLogger } from '@google/gemini-cli-core';
+import { exitCli } from '../utils.js';
+import { getLogFilePath } from './constants.js';
+
+export async function readLastLines(
+  filePath: string,
+  count: number,
+): Promise<string> {
+  if (count <= 0) {
+    return '';
+  }
+
+  const CHUNK_SIZE = 64 * 1024;
+  const fileHandle = await fs.promises.open(filePath, fs.constants.O_RDONLY);
+
+  try {
+    const stats = await fileHandle.stat();
+    if (stats.size === 0) {
+      return '';
+    }
+
+    const chunks: Buffer[] = [];
+    let totalBytes = 0;
+    let newlineCount = 0;
+    let position = stats.size;
+
+    while (position > 0 && newlineCount <= count) {
+      const readSize = Math.min(CHUNK_SIZE, position);
+      position -= readSize;
+
+      const buffer = Buffer.allocUnsafe(readSize);
+      const { bytesRead } = await fileHandle.read(
+        buffer,
+        0,
+        readSize,
+        position,
+      );
+
+      if (bytesRead === 0) {
+        break;
+      }
+
+      const chunk =
+        bytesRead === readSize ? buffer : buffer.subarray(0, bytesRead);
+      chunks.unshift(chunk);
+      totalBytes += chunk.length;
+
+      for (const byte of chunk) {
+        if (byte === 0x0a) {
+          newlineCount += 1;
+        }
+      }
+    }
+
+    const content = Buffer.concat(chunks, totalBytes).toString('utf-8');
+    const lines = content.split('\n');
+
+    if (position > 0 && lines.length > 0) {
+      const boundary = Buffer.allocUnsafe(1);
+      const { bytesRead } = await fileHandle.read(boundary, 0, 1, position - 1);
+      if (bytesRead === 1 && boundary[0] !== 0x0a) {
+        lines.shift();
+      }
+    }
+
+    if (lines.length > 0 && lines[lines.length - 1] === '') {
+      lines.pop();
+    }
+
+    if (lines.length === 0) {
+      return '';
+    }
+
+    return lines.slice(-count).join('\n') + '\n';
+  } finally {
+    await fileHandle.close();
+  }
+}
+
+interface LogsArgs {
+  lines?: number;
+  follow?: boolean;
+}
+
+function waitForChild(child: ChildProcess): Promise<number> {
+  return new Promise((resolve, reject) => {
+    child.once('error', reject);
+    child.once('close', (code) => resolve(code ?? 1));
+  });
+}
+
+async function runTail(logPath: string, lines: number, follow: boolean) {
+  const tailArgs = follow
+    ? ['-f', '-n', String(lines), logPath]
+    : ['-n', String(lines), logPath];
+  const child = spawn('tail', tailArgs, { stdio: 'inherit' });
+
+  if (!follow) {
+    return waitForChild(child);
+  }
+
+  const handleSigint = () => {
+    child.kill('SIGTERM');
+  };
+  process.once('SIGINT', handleSigint);
+
+  try {
+    return await waitForChild(child);
+  } finally {
+    process.off('SIGINT', handleSigint);
+  }
+}
+
+export const logsCommand: CommandModule<object, LogsArgs> = {
+  command: 'logs',
+  describe: 'View LiteRT-LM server logs',
+  builder: (yargs) =>
+    yargs
+      .option('lines', {
+        alias: 'n',
+        type: 'number',
+        description: 'Show the last N lines and exit (omit to follow live)',
+      })
+      .option('follow', {
+        alias: 'f',
+        type: 'boolean',
+        description:
+          'Follow log output (defaults to true when --lines is omitted)',
+      }),
+  handler: async (argv) => {
+    const logPath = getLogFilePath();
+
+    try {
+      await fs.promises.access(logPath, fs.constants.F_OK);
+    } catch {
+      debugLogger.log(`No log file found at ${logPath}`);
+      debugLogger.log(
+        'Is the LiteRT server running? Start it with: gemini gemma start',
+      );
+      await exitCli(1);
+      return;
+    }
+
+    const lines = argv.lines;
+    const follow = argv.follow ?? lines === undefined;
+    const requestedLines = lines ?? 20;
+
+    if (follow && process.platform === 'win32') {
+      debugLogger.log(
+        'Live log following is not supported on Windows. Use --lines N to view recent logs.',
+      );
+      await exitCli(1);
+      return;
+    }
+
+    if (process.platform === 'win32') {
+      process.stdout.write(await readLastLines(logPath, requestedLines));
+      await exitCli(0);
+      return;
+    }
+
+    try {
+      if (follow) {
+        debugLogger.log(`Tailing ${logPath} (Ctrl+C to stop)\n`);
+      }
+      const exitCode = await runTail(logPath, requestedLines, follow);
+      await exitCli(exitCode);
+    } catch (error) {
+      if (
+        error instanceof Error &&
+        'code' in error &&
+        error.code === 'ENOENT'
+      ) {
+        if (!follow) {
+          process.stdout.write(await readLastLines(logPath, requestedLines));
+          await exitCli(0);
+        } else {
+          debugLogger.error(
+            '"tail" command not found. Use --lines N to view recent logs without tail.',
+          );
+          await exitCli(1);
+        }
+      } else {
+        debugLogger.error(
+          `Failed to read log output: ${error instanceof Error ? error.message : String(error)}`,
+        );
+        await exitCli(1);
+      }
+    }
+  },
+};

packages/cli/src/commands/gemma/platform.test.ts

@@ -0,0 +1,162 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import fs from 'node:fs';
+import path from 'node:path';
+import { beforeEach, describe, expect, it, vi } from 'vitest';
+import { SettingScope } from '../../config/settings.js';
+import { getLiteRtBinDir } from './constants.js';
+
+const mockLoadSettings = vi.hoisted(() => vi.fn());
+
+vi.mock('../../config/settings.js', () => ({
+  loadSettings: mockLoadSettings,
+  SettingScope: {
+    User: 'User',
+  },
+}));
+
+import {
+  getBinaryPath,
+  isExpectedLiteRtServerCommand,
+  isBinaryInstalled,
+  readServerProcessInfo,
+  resolveGemmaConfig,
+} from './platform.js';
+
+describe('gemma platform helpers', () => {
+  function createMockSettings(
+    userGemmaSettings?: object,
+    mergedGemmaSettings?: object,
+  ) {
+    return {
+      merged: {
+        experimental: {
+          gemmaModelRouter: mergedGemmaSettings,
+        },
+      },
+      forScope: vi.fn((scope: SettingScope) => {
+        if (scope !== SettingScope.User) {
+          throw new Error(`Unexpected scope ${scope}`);
+        }
+        return {
+          settings: {
+            experimental: {
+              gemmaModelRouter: userGemmaSettings,
+            },
+          },
+        };
+      }),
+    };
+  }
+
+  beforeEach(() => {
+    vi.clearAllMocks();
+    mockLoadSettings.mockReturnValue(createMockSettings());
+  });
+
+  it('prefers the configured binary path from settings', () => {
+    mockLoadSettings.mockReturnValue(
+      createMockSettings({ binaryPath: '/custom/lit' }),
+    );
+
+    expect(getBinaryPath('lit.test')).toBe('/custom/lit');
+  });
+
+  it('ignores workspace overrides for the configured binary path', () => {
+    mockLoadSettings.mockReturnValue(
+      createMockSettings(
+        { binaryPath: '/user/lit' },
+        { binaryPath: '/workspace/evil' },
+      ),
+    );
+
+    expect(getBinaryPath('lit.test')).toBe('/user/lit');
+  });
+
+  it('falls back to the default install location when no custom path is set', () => {
+    expect(getBinaryPath('lit.test')).toBe(
+      path.join(getLiteRtBinDir(), 'lit.test'),
+    );
+  });
+
+  it('resolves the configured port and binary path from settings', () => {
+    mockLoadSettings.mockReturnValue(
+      createMockSettings(
+        { binaryPath: '/custom/lit' },
+        {
+          enabled: true,
+          classifier: {
+            host: 'http://localhost:8123/v1beta',
+          },
+        },
+      ),
+    );
+
+    expect(resolveGemmaConfig(9379)).toEqual({
+      settingsEnabled: true,
+      configuredPort: 8123,
+      configuredBinaryPath: '/custom/lit',
+    });
+  });
+
+  it('checks binary installation using the resolved binary path', () => {
+    mockLoadSettings.mockReturnValue(
+      createMockSettings({ binaryPath: '/custom/lit' }),
+    );
+    vi.spyOn(fs, 'existsSync').mockReturnValue(true);
+
+    expect(isBinaryInstalled()).toBe(true);
+    expect(fs.existsSync).toHaveBeenCalledWith('/custom/lit');
+  });
+
+  it('parses structured server process info from the pid file', () => {
+    vi.spyOn(fs, 'readFileSync').mockReturnValue(
+      JSON.stringify({
+        pid: 1234,
+        binaryPath: '/custom/lit',
+        port: 8123,
+      }),
+    );
+
+    expect(readServerProcessInfo()).toEqual({
+      pid: 1234,
+      binaryPath: '/custom/lit',
+      port: 8123,
+    });
+  });
+
+  it('parses legacy pid-only files for backward compatibility', () => {
+    vi.spyOn(fs, 'readFileSync').mockReturnValue('4321');
+
+    expect(readServerProcessInfo()).toEqual({
+      pid: 4321,
+    });
+  });
+
+  it('matches only the expected LiteRT serve command', () => {
+    expect(
+      isExpectedLiteRtServerCommand('/custom/lit serve --port=8123 --verbose', {
+        binaryPath: '/custom/lit',
+        port: 8123,
+      }),
+    ).toBe(true);
+
+    expect(
+      isExpectedLiteRtServerCommand('/custom/lit run --port=8123', {
+        binaryPath: '/custom/lit',
+        port: 8123,
+      }),
+    ).toBe(false);
+
+    expect(
+      isExpectedLiteRtServerCommand('/custom/lit serve --port=9000', {
+        binaryPath: '/custom/lit',
+        port: 8123,
+      }),
+    ).toBe(false);
+  });
+});

packages/cli/src/commands/gemma/platform.ts

@@ -0,0 +1,316 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import { loadSettings, SettingScope } from '../../config/settings.js';
+import fs from 'node:fs';
+import path from 'node:path';
+import { execFileSync } from 'node:child_process';
+import {
+  PLATFORM_BINARY_MAP,
+  LITERT_RELEASE_BASE_URL,
+  LITERT_RELEASE_VERSION,
+  getLiteRtBinDir,
+  GEMMA_MODEL_NAME,
+  HEALTH_CHECK_TIMEOUT_MS,
+  LITERT_API_VERSION,
+  getPidFilePath,
+} from './constants.js';
+
+export interface PlatformInfo {
+  key: string;
+  binaryName: string;
+}
+
+export interface GemmaConfigStatus {
+  settingsEnabled: boolean;
+  configuredPort: number;
+  configuredBinaryPath?: string;
+}
+
+export interface LiteRtServerProcessInfo {
+  pid: number;
+  binaryPath?: string;
+  port?: number;
+}
+
+function getUserConfiguredBinaryPath(
+  workspaceDir = process.cwd(),
+): string | undefined {
+  try {
+    const userGemmaSettings = loadSettings(workspaceDir).forScope(
+      SettingScope.User,
+    ).settings.experimental?.gemmaModelRouter;
+    return userGemmaSettings?.binaryPath?.trim() || undefined;
+  } catch {
+    return undefined;
+  }
+}
+
+function parsePortFromHost(
+  host: string | undefined,
+  fallbackPort: number,
+): number {
+  if (!host) {
+    return fallbackPort;
+  }
+
+  try {
+    const url = new URL(host);
+    const port = Number(url.port);
+    return Number.isFinite(port) && port > 0 ? port : fallbackPort;
+  } catch {
+    const match = host.match(/:(\d+)/);
+    if (!match) {
+      return fallbackPort;
+    }
+    const port = parseInt(match[1], 10);
+    return Number.isFinite(port) && port > 0 ? port : fallbackPort;
+  }
+}
+
+export function resolveGemmaConfig(fallbackPort: number): GemmaConfigStatus {
+  let settingsEnabled = false;
+  let configuredPort = fallbackPort;
+  const configuredBinaryPath = getUserConfiguredBinaryPath();
+  try {
+    const settings = loadSettings(process.cwd());
+    const gemmaSettings = settings.merged.experimental?.gemmaModelRouter;
+    settingsEnabled = gemmaSettings?.enabled === true;
+    configuredPort = parsePortFromHost(
+      gemmaSettings?.classifier?.host,
+      fallbackPort,
+    );
+  } catch {
+    // ignore — settings may fail to load outside a workspace
+  }
+  return { settingsEnabled, configuredPort, configuredBinaryPath };
+}
+
+export function detectPlatform(): PlatformInfo | null {
+  const key = `${process.platform}-${process.arch}`;
+  const binaryName = PLATFORM_BINARY_MAP[key];
+  if (!binaryName) {
+    return null;
+  }
+  return { key, binaryName };
+}
+
+export function getBinaryPath(binaryName?: string): string | null {
+  const configuredBinaryPath = getUserConfiguredBinaryPath();
+  if (configuredBinaryPath) {
+    return configuredBinaryPath;
+  }
+
+  const name = binaryName ?? detectPlatform()?.binaryName;
+  if (!name) return null;
+  return path.join(getLiteRtBinDir(), name);
+}
+
+export function getBinaryDownloadUrl(binaryName: string): string {
+  return `${LITERT_RELEASE_BASE_URL}/${LITERT_RELEASE_VERSION}/${binaryName}`;
+}
+
+export function isBinaryInstalled(binaryPath = getBinaryPath()): boolean {
+  if (!binaryPath) return false;
+  return fs.existsSync(binaryPath);
+}
+
+export function isModelDownloaded(binaryPath: string): boolean {
+  try {
+    const output = execFileSync(binaryPath, ['list'], {
+      encoding: 'utf-8',
+      timeout: 10000,
+    });
+    return output.includes(GEMMA_MODEL_NAME);
+  } catch {
+    return false;
+  }
+}
+
+export async function isServerRunning(port: number): Promise<boolean> {
+  try {
+    const controller = new AbortController();
+    const timeout = setTimeout(
+      () => controller.abort(),
+      HEALTH_CHECK_TIMEOUT_MS,
+    );
+    const response = await fetch(
+      `http://localhost:${port}/${LITERT_API_VERSION}/models/${GEMMA_MODEL_NAME}:generateContent`,
+      { method: 'POST', signal: controller.signal },
+    );
+    clearTimeout(timeout);
+    // A 400 (bad request) confirms the route exists — the server recognises
+    // the model endpoint.  Only a 404 means "wrong server / wrong model".
+    return response.status !== 404;
+  } catch {
+    return false;
+  }
+}
+
+function isLiteRtServerProcessInfo(
+  value: unknown,
+): value is LiteRtServerProcessInfo {
+  if (!value || typeof value !== 'object') {
+    return false;
+  }
+
+  const isPositiveInteger = (candidate: unknown): candidate is number =>
+    typeof candidate === 'number' &&
+    Number.isInteger(candidate) &&
+    candidate > 0;
+  const isNonEmptyString = (candidate: unknown): candidate is string =>
+    typeof candidate === 'string' && candidate.length > 0;
+
+  const pid: unknown = Object.getOwnPropertyDescriptor(value, 'pid')?.value;
+  if (!isPositiveInteger(pid)) {
+    return false;
+  }
+
+  const binaryPath: unknown = Object.getOwnPropertyDescriptor(
+    value,
+    'binaryPath',
+  )?.value;
+  if (binaryPath !== undefined && !isNonEmptyString(binaryPath)) {
+    return false;
+  }
+
+  const port: unknown = Object.getOwnPropertyDescriptor(value, 'port')?.value;
+  if (port !== undefined && !isPositiveInteger(port)) {
+    return false;
+  }
+
+  return true;
+}
+
+export function readServerProcessInfo(): LiteRtServerProcessInfo | null {
+  const pidPath = getPidFilePath();
+  try {
+    const content = fs.readFileSync(pidPath, 'utf-8').trim();
+    if (!content) {
+      return null;
+    }
+
+    if (/^\d+$/.test(content)) {
+      return { pid: parseInt(content, 10) };
+    }
+
+    const parsed = JSON.parse(content) as unknown;
+    return isLiteRtServerProcessInfo(parsed) ? parsed : null;
+  } catch {
+    return null;
+  }
+}
+
+export function writeServerProcessInfo(
+  processInfo: LiteRtServerProcessInfo,
+): void {
+  fs.writeFileSync(getPidFilePath(), JSON.stringify(processInfo), 'utf-8');
+}
+
+export function readServerPid(): number | null {
+  return readServerProcessInfo()?.pid ?? null;
+}
+
+function normalizeProcessValue(value: string): string {
+  const normalized = value.replace(/\0/g, ' ').trim();
+  if (process.platform === 'win32') {
+    return normalized.replace(/\\/g, '/').replace(/\s+/g, ' ').toLowerCase();
+  }
+  return normalized.replace(/\s+/g, ' ');
+}
+
+function readProcessCommandLine(pid: number): string | null {
+  try {
+    if (process.platform === 'linux') {
+      const output = fs.readFileSync(`/proc/${pid}/cmdline`, 'utf-8');
+      return output.trim() ? output : null;
+    }
+
+    if (process.platform === 'win32') {
+      const output = execFileSync(
+        'powershell.exe',
+        [
+          '-NoProfile',
+          '-Command',
+          `(Get-CimInstance Win32_Process -Filter "ProcessId = ${pid}").CommandLine`,
+        ],
+        {
+          encoding: 'utf-8',
+          timeout: 5000,
+        },
+      );
+      return output.trim() || null;
+    }
+
+    const output = execFileSync('ps', ['-p', String(pid), '-o', 'command='], {
+      encoding: 'utf-8',
+      timeout: 5000,
+    });
+    return output.trim() || null;
+  } catch {
+    return null;
+  }
+}
+
+export function isExpectedLiteRtServerCommand(
+  commandLine: string,
+  options: {
+    binaryPath?: string | null;
+    port?: number;
+  },
+): boolean {
+  const normalizedCommandLine = normalizeProcessValue(commandLine);
+  if (!normalizedCommandLine) {
+    return false;
+  }
+
+  if (!/(^|\s|")serve(\s|$)/.test(normalizedCommandLine)) {
+    return false;
+  }
+
+  if (
+    options.port !== undefined &&
+    !normalizedCommandLine.includes(`--port=${options.port}`)
+  ) {
+    return false;
+  }
+
+  if (!options.binaryPath) {
+    return true;
+  }
+
+  const normalizedBinaryPath = normalizeProcessValue(options.binaryPath);
+  const normalizedBinaryName = normalizeProcessValue(
+    path.basename(options.binaryPath),
+  );
+  return (
+    normalizedCommandLine.includes(normalizedBinaryPath) ||
+    normalizedCommandLine.includes(normalizedBinaryName)
+  );
+}
+
+export function isExpectedLiteRtServerProcess(
+  pid: number,
+  options: {
+    binaryPath?: string | null;
+    port?: number;
+  },
+): boolean {
+  const commandLine = readProcessCommandLine(pid);
+  if (!commandLine) {
+    return false;
+  }
+  return isExpectedLiteRtServerCommand(commandLine, options);
+}
+
+export function isProcessRunning(pid: number): boolean {
+  try {
+    process.kill(pid, 0);
+    return true;
+  } catch {
+    return false;
+  }
+}

packages/cli/src/commands/gemma/setup.test.ts

@@ -0,0 +1,60 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import fs from 'node:fs';
+import os from 'node:os';
+import path from 'node:path';
+import { afterEach, describe, expect, it } from 'vitest';
+import { PLATFORM_BINARY_MAP, PLATFORM_BINARY_SHA256 } from './constants.js';
+import { computeFileSha256, verifyFileSha256 } from './setup.js';
+
+describe('gemma setup checksum helpers', () => {
+  const tempFiles: string[] = [];
+
+  afterEach(async () => {
+    await Promise.all(
+      tempFiles
+        .splice(0)
+        .map((filePath) => fs.promises.rm(filePath, { force: true })),
+    );
+  });
+
+  it('has a pinned checksum for every supported LiteRT binary', () => {
+    expect(Object.keys(PLATFORM_BINARY_SHA256).sort()).toEqual(
+      Object.values(PLATFORM_BINARY_MAP).sort(),
+    );
+  });
+
+  it('computes the sha256 for a downloaded file', async () => {
+    const filePath = path.join(
+      os.tmpdir(),
+      `gemma-setup-${Date.now()}-${Math.random().toString(36).slice(2)}`,
+    );
+    tempFiles.push(filePath);
+    await fs.promises.writeFile(filePath, 'hello world', 'utf-8');
+
+    await expect(computeFileSha256(filePath)).resolves.toBe(
+      'b94d27b9934d3e08a52e52d7da7dabfac484efe37a5380ee9088f7ace2efcde9',
+    );
+  });
+
+  it('verifies whether a file matches the expected sha256', async () => {
+    const filePath = path.join(
+      os.tmpdir(),
+      `gemma-setup-${Date.now()}-${Math.random().toString(36).slice(2)}`,
+    );
+    tempFiles.push(filePath);
+    await fs.promises.writeFile(filePath, 'hello world', 'utf-8');
+
+    await expect(
+      verifyFileSha256(
+        filePath,
+        'b94d27b9934d3e08a52e52d7da7dabfac484efe37a5380ee9088f7ace2efcde9',
+      ),
+    ).resolves.toBe(true);
+    await expect(verifyFileSha256(filePath, 'deadbeef')).resolves.toBe(false);
+  });
+});

packages/cli/src/commands/gemma/setup.ts

@@ -0,0 +1,504 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { CommandModule } from 'yargs';
+import { createHash } from 'node:crypto';
+import fs from 'node:fs';
+import path from 'node:path';
+import { execFileSync, spawn as nodeSpawn } from 'node:child_process';
+import chalk from 'chalk';
+import { debugLogger } from '@google/gemini-cli-core';
+import { loadSettings, SettingScope } from '../../config/settings.js';
+import { exitCli } from '../utils.js';
+import {
+  DEFAULT_PORT,
+  GEMMA_MODEL_NAME,
+  PLATFORM_BINARY_SHA256,
+} from './constants.js';
+import {
+  detectPlatform,
+  getBinaryDownloadUrl,
+  getBinaryPath,
+  isBinaryInstalled,
+  isModelDownloaded,
+} from './platform.js';
+import { startServer } from './start.js';
+import readline from 'node:readline';
+
+const log = (msg: string) => debugLogger.log(msg);
+const logError = (msg: string) => debugLogger.error(msg);
+
+async function promptYesNo(question: string): Promise<boolean> {
+  const rl = readline.createInterface({
+    input: process.stdin,
+    output: process.stdout,
+  });
+  return new Promise((resolve) => {
+    rl.question(`${question} (y/N): `, (answer) => {
+      rl.close();
+      resolve(
+        answer.trim().toLowerCase() === 'y' ||
+          answer.trim().toLowerCase() === 'yes',
+      );
+    });
+  });
+}
+
+function formatBytes(bytes: number): string {
+  if (bytes < 1024) return `${bytes} B`;
+  if (bytes < 1024 * 1024) return `${(bytes / 1024).toFixed(1)} KB`;
+  return `${(bytes / (1024 * 1024)).toFixed(1)} MB`;
+}
+
+function renderProgress(downloaded: number, total: number | null): void {
+  const barWidth = 30;
+  if (total && total > 0) {
+    const pct = Math.min(downloaded / total, 1);
+    const filled = Math.round(barWidth * pct);
+    const bar = '█'.repeat(filled) + '░'.repeat(barWidth - filled);
+    const pctStr = (pct * 100).toFixed(0).padStart(3);
+    process.stderr.write(
+      `\r  [${bar}] ${pctStr}% ${formatBytes(downloaded)} / ${formatBytes(total)}`,
+    );
+  } else {
+    process.stderr.write(`\r  Downloaded ${formatBytes(downloaded)}`);
+  }
+}
+
+async function downloadFile(url: string, destPath: string): Promise<void> {
+  const tmpPath = destPath + '.downloading';
+  if (fs.existsSync(tmpPath)) {
+    fs.unlinkSync(tmpPath);
+  }
+
+  const response = await fetch(url, { redirect: 'follow' });
+  if (!response.ok) {
+    throw new Error(
+      `Download failed: HTTP ${response.status} ${response.statusText}`,
+    );
+  }
+  if (!response.body) {
+    throw new Error('Download failed: No response body');
+  }
+
+  const contentLength = response.headers.get('content-length');
+  const totalBytes = contentLength ? parseInt(contentLength, 10) : null;
+  let downloadedBytes = 0;
+
+  const fileStream = fs.createWriteStream(tmpPath);
+  const reader = response.body.getReader();
+
+  try {
+    for (;;) {
+      const { done, value } = await reader.read();
+      if (done) break;
+      const writeOk = fileStream.write(value);
+      if (!writeOk) {
+        await new Promise<void>((resolve) => fileStream.once('drain', resolve));
+      }
+      downloadedBytes += value.byteLength;
+      renderProgress(downloadedBytes, totalBytes);
+    }
+  } finally {
+    fileStream.end();
+    process.stderr.write('\r' + ' '.repeat(80) + '\r');
+  }
+
+  await new Promise<void>((resolve, reject) => {
+    fileStream.on('finish', resolve);
+    fileStream.on('error', reject);
+  });
+
+  fs.renameSync(tmpPath, destPath);
+}
+
+export async function computeFileSha256(filePath: string): Promise<string> {
+  const hash = createHash('sha256');
+  const fileStream = fs.createReadStream(filePath);
+
+  return new Promise((resolve, reject) => {
+    fileStream.on('data', (chunk) => {
+      hash.update(chunk);
+    });
+    fileStream.on('error', reject);
+    fileStream.on('end', () => {
+      resolve(hash.digest('hex'));
+    });
+  });
+}
+
+export async function verifyFileSha256(
+  filePath: string,
+  expectedHash: string,
+): Promise<boolean> {
+  const actualHash = await computeFileSha256(filePath);
+  return actualHash === expectedHash;
+}
+
+function spawnInherited(command: string, args: string[]): Promise<number> {
+  return new Promise((resolve, reject) => {
+    const child = nodeSpawn(command, args, {
+      stdio: 'inherit',
+    });
+    child.on('close', (code) => resolve(code ?? 1));
+    child.on('error', reject);
+  });
+}
+
+interface SetupArgs {
+  port: number;
+  skipModel: boolean;
+  start: boolean;
+  force: boolean;
+  consent: boolean;
+}
+
+async function handleSetup(argv: SetupArgs): Promise<number> {
+  const { port, force } = argv;
+  let settingsUpdated = false;
+  let serverStarted = false;
+  let autoStartServer = true;
+
+  log('');
+  log(chalk.bold('Gemma Local Model Routing Setup'));
+  log(chalk.dim('─'.repeat(40)));
+  log('');
+
+  const platform = detectPlatform();
+  if (!platform) {
+    logError(
+      chalk.red(`Unsupported platform: ${process.platform}-${process.arch}`),
+    );
+    logError(
+      'LiteRT-LM binaries are available for: macOS (ARM64), Linux (x86_64), Windows (x86_64)',
+    );
+    return 1;
+  }
+  log(chalk.dim(`  Platform: ${platform.key} → ${platform.binaryName}`));
+
+  if (!argv.consent) {
+    log('');
+    log('This will download and install the LiteRT-LM runtime and the');
+    log(
+      `Gemma model (${GEMMA_MODEL_NAME}, ~1 GB). By proceeding, you agree to the`,
+    );
+    log('Gemma Terms of Use: https://ai.google.dev/gemma/terms');
+    log('');
+
+    const accepted = await promptYesNo('Do you want to continue?');
+    if (!accepted) {
+      log('Setup cancelled.');
+      return 0;
+    }
+  }
+
+  const binaryPath = getBinaryPath(platform.binaryName)!;
+  const alreadyInstalled = isBinaryInstalled();
+
+  if (alreadyInstalled && !force) {
+    log('');
+    log(chalk.green('  ✓ LiteRT-LM binary already installed at:'));
+    log(chalk.dim(`    ${binaryPath}`));
+  } else {
+    log('');
+    log('  Downloading LiteRT-LM binary...');
+    const downloadUrl = getBinaryDownloadUrl(platform.binaryName);
+    debugLogger.log(`Downloading from: ${downloadUrl}`);
+
+    try {
+      const binDir = path.dirname(binaryPath);
+      fs.mkdirSync(binDir, { recursive: true });
+      await downloadFile(downloadUrl, binaryPath);
+      log(chalk.green('  ✓ Binary downloaded successfully'));
+    } catch (error) {
+      logError(
+        chalk.red(
+          `  ✗ Failed to download binary: ${error instanceof Error ? error.message : String(error)}`,
+        ),
+      );
+      logError('  Check your internet connection and try again.');
+      return 1;
+    }
+
+    const expectedHash = PLATFORM_BINARY_SHA256[platform.binaryName];
+    if (!expectedHash) {
+      logError(
+        chalk.red(
+          `  ✗ No checksum is configured for ${platform.binaryName}. Refusing to install the binary.`,
+        ),
+      );
+      try {
+        fs.rmSync(binaryPath, { force: true });
+      } catch {
+        // ignore
+      }
+      return 1;
+    }
+
+    try {
+      const checksumVerified = await verifyFileSha256(binaryPath, expectedHash);
+      if (!checksumVerified) {
+        logError(
+          chalk.red(
+            '  ✗ Downloaded binary checksum did not match the expected release hash.',
+          ),
+        );
+        try {
+          fs.rmSync(binaryPath, { force: true });
+        } catch {
+          // ignore
+        }
+        return 1;
+      }
+      log(chalk.green('  ✓ Binary checksum verified'));
+    } catch (error) {
+      logError(
+        chalk.red(
+          `  ✗ Failed to verify binary checksum: ${error instanceof Error ? error.message : String(error)}`,
+        ),
+      );
+      try {
+        fs.rmSync(binaryPath, { force: true });
+      } catch {
+        // ignore
+      }
+      return 1;
+    }
+
+    if (process.platform !== 'win32') {
+      try {
+        fs.chmodSync(binaryPath, 0o755);
+      } catch (error) {
+        logError(
+          chalk.red(
+            `  ✗ Failed to set executable permission: ${error instanceof Error ? error.message : String(error)}`,
+          ),
+        );
+        return 1;
+      }
+    }
+
+    if (process.platform === 'darwin') {
+      try {
+        execFileSync('xattr', ['-d', 'com.apple.quarantine', binaryPath], {
+          stdio: 'ignore',
+        });
+        log(chalk.green('  ✓ macOS quarantine attribute removed'));
+      } catch {
+        // Expected if the attribute doesn't exist.
+      }
+    }
+  }
+
+  if (!argv.skipModel) {
+    const modelAlreadyDownloaded = isModelDownloaded(binaryPath);
+    if (modelAlreadyDownloaded && !force) {
+      log('');
+      log(chalk.green(`  ✓ Model ${GEMMA_MODEL_NAME} already downloaded`));
+    } else {
+      log('');
+      log(`  Downloading model ${GEMMA_MODEL_NAME}...`);
+      log(chalk.dim('  You may be prompted to accept the Gemma Terms of Use.'));
+      log('');
+
+      const exitCode = await spawnInherited(binaryPath, [
+        'pull',
+        GEMMA_MODEL_NAME,
+      ]);
+      if (exitCode !== 0) {
+        logError('');
+        logError(
+          chalk.red(`  ✗ Model download failed (exit code ${exitCode})`),
+        );
+        return 1;
+      }
+      log('');
+      log(chalk.green(`  ✓ Model ${GEMMA_MODEL_NAME} downloaded`));
+    }
+  }
+
+  log('');
+  log('  Configuring settings...');
+  try {
+    const settings = loadSettings(process.cwd());
+
+    // User scope: security-sensitive settings that must not be overridable
+    // by workspace configs (prevents arbitrary binary execution).
+    const existingUserGemma =
+      settings.forScope(SettingScope.User).settings.experimental
+        ?.gemmaModelRouter ?? {};
+    autoStartServer = existingUserGemma.autoStartServer ?? true;
+    const existingUserExperimental =
+      settings.forScope(SettingScope.User).settings.experimental ?? {};
+    settings.setValue(SettingScope.User, 'experimental', {
+      ...existingUserExperimental,
+      gemmaModelRouter: {
+        autoStartServer,
+        ...(existingUserGemma.binaryPath !== undefined
+          ? { binaryPath: existingUserGemma.binaryPath }
+          : {}),
+      },
+    });
+
+    // Workspace scope: project-isolated settings so the local model only
+    // runs for this specific project, saving resources globally.
+    const existingWorkspaceGemma =
+      settings.forScope(SettingScope.Workspace).settings.experimental
+        ?.gemmaModelRouter ?? {};
+    const existingWorkspaceExperimental =
+      settings.forScope(SettingScope.Workspace).settings.experimental ?? {};
+    settings.setValue(SettingScope.Workspace, 'experimental', {
+      ...existingWorkspaceExperimental,
+      gemmaModelRouter: {
+        ...existingWorkspaceGemma,
+        enabled: true,
+        classifier: {
+          ...existingWorkspaceGemma.classifier,
+          host: `http://localhost:${port}`,
+          model: GEMMA_MODEL_NAME,
+        },
+      },
+    });
+
+    log(chalk.green('  ✓ Settings updated'));
+    log(chalk.dim('    User (~/.gemini/settings.json): autoStartServer'));
+    log(
+      chalk.dim('    Workspace (.gemini/settings.json): enabled, classifier'),
+    );
+    settingsUpdated = true;
+  } catch (error) {
+    logError(
+      chalk.red(
+        `  ✗ Failed to update settings: ${error instanceof Error ? error.message : String(error)}`,
+      ),
+    );
+    logError(
+      '  You can manually add the configuration to ~/.gemini/settings.json',
+    );
+  }
+
+  if (argv.start) {
+    log('');
+    log('  Starting LiteRT server...');
+    serverStarted = await startServer(binaryPath, port);
+    if (serverStarted) {
+      log(chalk.green(`  ✓ Server started on port ${port}`));
+    } else {
+      log(
+        chalk.yellow(
+          `  ! Server may not have started correctly. Check: gemini gemma status`,
+        ),
+      );
+    }
+  }
+
+  const routingActive = settingsUpdated && serverStarted;
+  const setupSucceeded = settingsUpdated && (!argv.start || serverStarted);
+  log('');
+  log(chalk.dim('─'.repeat(40)));
+  if (routingActive) {
+    log(chalk.bold.green('  Setup complete! Local model routing is active.'));
+  } else if (settingsUpdated) {
+    log(
+      chalk.bold.green('  Setup complete! Local model routing is configured.'),
+    );
+  } else {
+    log(
+      chalk.bold.yellow(
+        '  Setup incomplete. Manual settings changes are still required.',
+      ),
+    );
+  }
+  log('');
+  log('  How it works: Every request is classified by the local Gemma model.');
+  log(
+    '  Simple tasks (file reads, quick edits) route to ' +
+      chalk.cyan('Flash') +
+      ' for speed.',
+  );
+  log(
+    '  Complex tasks (debugging, architecture) route to ' +
+      chalk.cyan('Pro') +
+      ' for quality.',
+  );
+  log('  This happens automatically — just use the CLI as usual.');
+  log('');
+  if (!settingsUpdated) {
+    log(
+      chalk.yellow(
+        '  Fix the settings update above, then rerun "gemini gemma status".',
+      ),
+    );
+    log('');
+  } else if (!argv.start) {
+    log(chalk.yellow('  Note: Run "gemini gemma start" to start the server.'));
+    if (autoStartServer) {
+      log(
+        chalk.yellow(
+          '  Or restart the CLI to auto-start it on the next launch.',
+        ),
+      );
+    }
+    log('');
+  } else if (!serverStarted) {
+    log(
+      chalk.yellow(
+        '  Review the server logs and rerun "gemini gemma start" after fixing the issue.',
+      ),
+    );
+    log('');
+  }
+  log('  Useful commands:');
+  log(chalk.dim('    gemini gemma status   Check routing status'));
+  log(chalk.dim('    gemini gemma start    Start the LiteRT server'));
+  log(chalk.dim('    gemini gemma stop     Stop the LiteRT server'));
+  log(chalk.dim('    /gemma               Check status inside a session'));
+  log('');
+
+  return setupSucceeded ? 0 : 1;
+}
+
+export const setupCommand: CommandModule = {
+  command: 'setup',
+  describe: 'Download and configure Gemma local model routing',
+  builder: (yargs) =>
+    yargs
+      .option('port', {
+        type: 'number',
+        default: DEFAULT_PORT,
+        description: 'Port for the LiteRT server',
+      })
+      .option('skip-model', {
+        type: 'boolean',
+        default: false,
+        description: 'Skip model download (binary only)',
+      })
+      .option('start', {
+        type: 'boolean',
+        default: true,
+        description: 'Start the server after setup',
+      })
+      .option('force', {
+        type: 'boolean',
+        default: false,
+        description: 'Re-download binary and model even if already present',
+      })
+      .option('consent', {
+        type: 'boolean',
+        default: false,
+        description: 'Skip interactive consent prompt (implies acceptance)',
+      }),
+  handler: async (argv) => {
+    const exitCode = await handleSetup({
+      port: Number(argv['port']),
+      skipModel: Boolean(argv['skipModel']),
+      start: Boolean(argv['start']),
+      force: Boolean(argv['force']),
+      consent: Boolean(argv['consent']),
+    });
+    await exitCli(exitCode);
+  },
+};

packages/cli/src/commands/gemma/start.ts

@@ -0,0 +1,123 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { CommandModule } from 'yargs';
+import fs from 'node:fs';
+import path from 'node:path';
+import { spawn } from 'node:child_process';
+import chalk from 'chalk';
+import { debugLogger } from '@google/gemini-cli-core';
+import { exitCli } from '../utils.js';
+import {
+  DEFAULT_PORT,
+  getPidFilePath,
+  getLogFilePath,
+  getLiteRtBinDir,
+  SERVER_START_WAIT_MS,
+} from './constants.js';
+import {
+  getBinaryPath,
+  isBinaryInstalled,
+  isServerRunning,
+  resolveGemmaConfig,
+  writeServerProcessInfo,
+} from './platform.js';
+
+export async function startServer(
+  binaryPath: string,
+  port: number,
+): Promise<boolean> {
+  const alreadyRunning = await isServerRunning(port);
+  if (alreadyRunning) {
+    debugLogger.log(`LiteRT server already running on port ${port}`);
+    return true;
+  }
+
+  const logPath = getLogFilePath();
+  fs.mkdirSync(getLiteRtBinDir(), { recursive: true });
+  const tmpDir = path.dirname(getPidFilePath());
+  fs.mkdirSync(tmpDir, { recursive: true });
+
+  const logFd = fs.openSync(logPath, 'a');
+
+  try {
+    const child = spawn(binaryPath, ['serve', `--port=${port}`, '--verbose'], {
+      detached: true,
+      stdio: ['ignore', logFd, logFd],
+    });
+
+    if (child.pid) {
+      writeServerProcessInfo({
+        pid: child.pid,
+        binaryPath,
+        port,
+      });
+    }
+
+    child.unref();
+  } finally {
+    fs.closeSync(logFd);
+  }
+
+  await new Promise((resolve) => setTimeout(resolve, SERVER_START_WAIT_MS));
+  return isServerRunning(port);
+}
+
+export const startCommand: CommandModule = {
+  command: 'start',
+  describe: 'Start the LiteRT-LM server',
+  builder: (yargs) =>
+    yargs.option('port', {
+      type: 'number',
+      description: 'Port for the LiteRT server',
+    }),
+  handler: async (argv) => {
+    let port: number | undefined;
+    if (argv['port'] !== undefined) {
+      port = Number(argv['port']);
+    }
+
+    if (!port) {
+      const { configuredPort } = resolveGemmaConfig(DEFAULT_PORT);
+      port = configuredPort;
+    }
+
+    const binaryPath = getBinaryPath();
+    if (!binaryPath || !isBinaryInstalled(binaryPath)) {
+      debugLogger.error(
+        chalk.red(
+          'LiteRT-LM binary not found. Run "gemini gemma setup" first.',
+        ),
+      );
+      await exitCli(1);
+      return;
+    }
+
+    const alreadyRunning = await isServerRunning(port);
+    if (alreadyRunning) {
+      debugLogger.log(
+        chalk.green(`LiteRT server is already running on port ${port}.`),
+      );
+      await exitCli(0);
+      return;
+    }
+
+    debugLogger.log(`Starting LiteRT server on port ${port}...`);
+
+    const started = await startServer(binaryPath, port);
+    if (started) {
+      debugLogger.log(chalk.green(`LiteRT server started on port ${port}.`));
+      debugLogger.log(chalk.dim(`Logs: ${getLogFilePath()}`));
+      await exitCli(0);
+    } else {
+      debugLogger.error(
+        chalk.red('Server may not have started correctly. Check logs:'),
+      );
+      debugLogger.error(chalk.dim(`  ${getLogFilePath()}`));
+      await exitCli(1);
+    }
+  },
+};

packages/cli/src/commands/gemma/status.ts

@@ -0,0 +1,165 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { CommandModule } from 'yargs';
+import chalk from 'chalk';
+import { DEFAULT_PORT, GEMMA_MODEL_NAME } from './constants.js';
+import {
+  detectPlatform,
+  getBinaryPath,
+  isBinaryInstalled,
+  isModelDownloaded,
+  isServerRunning,
+  readServerPid,
+  isProcessRunning,
+  resolveGemmaConfig,
+} from './platform.js';
+import { exitCli } from '../utils.js';
+
+export interface GemmaStatusResult {
+  binaryInstalled: boolean;
+  binaryPath: string | null;
+  modelDownloaded: boolean;
+  serverRunning: boolean;
+  serverPid: number | null;
+  settingsEnabled: boolean;
+  port: number;
+  allPassing: boolean;
+}
+
+export async function checkGemmaStatus(
+  port?: number,
+): Promise<GemmaStatusResult> {
+  const { settingsEnabled, configuredPort } = resolveGemmaConfig(DEFAULT_PORT);
+
+  const effectivePort = port ?? configuredPort;
+  const binaryPath = getBinaryPath();
+  const binaryInstalled = isBinaryInstalled(binaryPath);
+  const modelDownloaded =
+    binaryInstalled && binaryPath ? isModelDownloaded(binaryPath) : false;
+  const serverRunning = await isServerRunning(effectivePort);
+  const pid = readServerPid();
+  const serverPid = pid && isProcessRunning(pid) ? pid : null;
+
+  const allPassing =
+    binaryInstalled && modelDownloaded && serverRunning && settingsEnabled;
+
+  return {
+    binaryInstalled,
+    binaryPath,
+    modelDownloaded,
+    serverRunning,
+    serverPid,
+    settingsEnabled,
+    port: effectivePort,
+    allPassing,
+  };
+}
+
+export function formatGemmaStatus(status: GemmaStatusResult): string {
+  const check = (ok: boolean) => (ok ? chalk.green('✓') : chalk.red('✗'));
+
+  const lines: string[] = [
+    '',
+    chalk.bold('Gemma Local Model Routing Status'),
+    chalk.dim('─'.repeat(40)),
+    '',
+  ];
+
+  if (status.binaryInstalled) {
+    lines.push(`  Binary:    ${check(true)} Installed (${status.binaryPath})`);
+  } else {
+    const platform = detectPlatform();
+    if (platform) {
+      lines.push(`  Binary:    ${check(false)} Not installed`);
+      lines.push(chalk.dim(`             Run: gemini gemma setup`));
+    } else {
+      lines.push(
+        `  Binary:    ${check(false)} Unsupported platform (${process.platform}-${process.arch})`,
+      );
+    }
+  }
+
+  if (status.modelDownloaded) {
+    lines.push(`  Model:     ${check(true)} ${GEMMA_MODEL_NAME} downloaded`);
+  } else {
+    lines.push(`  Model:     ${check(false)} ${GEMMA_MODEL_NAME} not found`);
+    if (status.binaryInstalled) {
+      lines.push(
+        chalk.dim(
+          `             Run: ${status.binaryPath} pull ${GEMMA_MODEL_NAME}`,
+        ),
+      );
+    } else {
+      lines.push(chalk.dim(`             Run: gemini gemma setup`));
+    }
+  }
+
+  if (status.serverRunning) {
+    const pidInfo = status.serverPid ? ` (PID ${status.serverPid})` : '';
+    lines.push(
+      `  Server:    ${check(true)} Running on port ${status.port}${pidInfo}`,
+    );
+  } else {
+    lines.push(
+      `  Server:    ${check(false)} Not running on port ${status.port}`,
+    );
+    lines.push(chalk.dim(`             Run: gemini gemma start`));
+  }
+
+  if (status.settingsEnabled) {
+    lines.push(`  Settings:  ${check(true)} Enabled in settings.json`);
+  } else {
+    lines.push(`  Settings:  ${check(false)} Not enabled in settings.json`);
+    lines.push(
+      chalk.dim(
+        `             Run: gemini gemma setup (auto-configures settings)`,
+      ),
+    );
+  }
+
+  lines.push('');
+
+  if (status.allPassing) {
+    lines.push(chalk.green('  Routing is active — no action needed.'));
+    lines.push('');
+    lines.push(
+      chalk.dim(
+        '  Simple requests → Flash (fast) | Complex requests → Pro (powerful)',
+      ),
+    );
+    lines.push(chalk.dim('  This happens automatically on every request.'));
+  } else {
+    lines.push(
+      chalk.yellow(
+        '  Some checks failed. Run "gemini gemma setup" for guided installation.',
+      ),
+    );
+  }
+
+  lines.push('');
+  return lines.join('\n');
+}
+
+export const statusCommand: CommandModule = {
+  command: 'status',
+  describe: 'Check Gemma local model routing status',
+  builder: (yargs) =>
+    yargs.option('port', {
+      type: 'number',
+      description: 'Port to check for the LiteRT server',
+    }),
+  handler: async (argv) => {
+    let port: number | undefined;
+    if (argv['port'] !== undefined) {
+      port = Number(argv['port']);
+    }
+    const status = await checkGemmaStatus(port);
+    const output = formatGemmaStatus(status);
+    process.stdout.write(output);
+    await exitCli(status.allPassing ? 0 : 1);
+  },
+};

packages/cli/src/commands/gemma/stop.test.ts

@@ -0,0 +1,112 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import fs from 'node:fs';
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
+
+const mockGetBinaryPath = vi.hoisted(() => vi.fn());
+const mockIsExpectedLiteRtServerProcess = vi.hoisted(() => vi.fn());
+const mockIsProcessRunning = vi.hoisted(() => vi.fn());
+const mockIsServerRunning = vi.hoisted(() => vi.fn());
+const mockReadServerPid = vi.hoisted(() => vi.fn());
+const mockReadServerProcessInfo = vi.hoisted(() => vi.fn());
+const mockResolveGemmaConfig = vi.hoisted(() => vi.fn());
+
+vi.mock('@google/gemini-cli-core', async (importOriginal) => {
+  const { mockCoreDebugLogger } = await import(
+    '../../test-utils/mockDebugLogger.js'
+  );
+  return mockCoreDebugLogger(
+    await importOriginal<typeof import('@google/gemini-cli-core')>(),
+    {
+      stripAnsi: false,
+    },
+  );
+});
+
+vi.mock('./constants.js', () => ({
+  DEFAULT_PORT: 9379,
+  getPidFilePath: vi.fn(() => '/tmp/litert-server.pid'),
+}));
+
+vi.mock('./platform.js', () => ({
+  getBinaryPath: mockGetBinaryPath,
+  isExpectedLiteRtServerProcess: mockIsExpectedLiteRtServerProcess,
+  isProcessRunning: mockIsProcessRunning,
+  isServerRunning: mockIsServerRunning,
+  readServerPid: mockReadServerPid,
+  readServerProcessInfo: mockReadServerProcessInfo,
+  resolveGemmaConfig: mockResolveGemmaConfig,
+}));
+
+vi.mock('../utils.js', () => ({
+  exitCli: vi.fn(),
+}));
+
+import { stopServer } from './stop.js';
+
+describe('gemma stop command', () => {
+  beforeEach(() => {
+    vi.clearAllMocks();
+    vi.useFakeTimers();
+    mockGetBinaryPath.mockReturnValue('/custom/lit');
+    mockResolveGemmaConfig.mockReturnValue({ configuredPort: 9379 });
+  });
+
+  afterEach(() => {
+    vi.useRealTimers();
+    vi.restoreAllMocks();
+  });
+
+  it('refuses to signal a pid that does not match the expected LiteRT server', async () => {
+    mockReadServerProcessInfo.mockReturnValue({
+      pid: 1234,
+      binaryPath: '/custom/lit',
+      port: 8123,
+    });
+    mockIsProcessRunning.mockReturnValue(true);
+    mockIsExpectedLiteRtServerProcess.mockReturnValue(false);
+
+    const killSpy = vi.spyOn(process, 'kill').mockImplementation(() => true);
+
+    await expect(stopServer(8123)).resolves.toBe('unexpected-process');
+    expect(killSpy).not.toHaveBeenCalled();
+  });
+
+  it('stops the verified LiteRT server and removes the pid file', async () => {
+    mockReadServerProcessInfo.mockReturnValue({
+      pid: 1234,
+      binaryPath: '/custom/lit',
+      port: 8123,
+    });
+    mockIsProcessRunning.mockReturnValueOnce(true).mockReturnValueOnce(false);
+    mockIsExpectedLiteRtServerProcess.mockReturnValue(true);
+
+    const unlinkSpy = vi.spyOn(fs, 'unlinkSync').mockImplementation(() => {});
+    const killSpy = vi.spyOn(process, 'kill').mockImplementation(() => true);
+
+    const stopPromise = stopServer(8123);
+    await vi.runAllTimersAsync();
+
+    await expect(stopPromise).resolves.toBe('stopped');
+    expect(killSpy).toHaveBeenCalledWith(1234, 'SIGTERM');
+    expect(unlinkSpy).toHaveBeenCalledWith('/tmp/litert-server.pid');
+  });
+
+  it('cleans up a stale pid file when the recorded process is no longer running', async () => {
+    mockReadServerProcessInfo.mockReturnValue({
+      pid: 1234,
+      binaryPath: '/custom/lit',
+      port: 8123,
+    });
+    mockIsProcessRunning.mockReturnValue(false);
+
+    const unlinkSpy = vi.spyOn(fs, 'unlinkSync').mockImplementation(() => {});
+
+    await expect(stopServer(8123)).resolves.toBe('not-running');
+    expect(unlinkSpy).toHaveBeenCalledWith('/tmp/litert-server.pid');
+  });
+});

packages/cli/src/commands/gemma/stop.ts

@@ -0,0 +1,155 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { CommandModule } from 'yargs';
+import fs from 'node:fs';
+import chalk from 'chalk';
+import { debugLogger } from '@google/gemini-cli-core';
+import { exitCli } from '../utils.js';
+import { DEFAULT_PORT, getPidFilePath } from './constants.js';
+import {
+  getBinaryPath,
+  isExpectedLiteRtServerProcess,
+  isProcessRunning,
+  isServerRunning,
+  readServerPid,
+  readServerProcessInfo,
+  resolveGemmaConfig,
+} from './platform.js';
+
+export type StopServerResult =
+  | 'stopped'
+  | 'not-running'
+  | 'unexpected-process'
+  | 'failed';
+
+export async function stopServer(
+  expectedPort?: number,
+): Promise<StopServerResult> {
+  const processInfo = readServerProcessInfo();
+  const pidPath = getPidFilePath();
+
+  if (!processInfo) {
+    return 'not-running';
+  }
+
+  const { pid } = processInfo;
+  if (!isProcessRunning(pid)) {
+    debugLogger.log(
+      `Stale PID file found (PID ${pid} is not running), removing ${pidPath}`,
+    );
+    try {
+      fs.unlinkSync(pidPath);
+    } catch {
+      // ignore
+    }
+    return 'not-running';
+  }
+
+  const binaryPath = processInfo.binaryPath ?? getBinaryPath();
+  const port = processInfo.port ?? expectedPort;
+  if (!isExpectedLiteRtServerProcess(pid, { binaryPath, port })) {
+    debugLogger.warn(
+      `Refusing to stop PID ${pid} because it does not match the expected LiteRT server process.`,
+    );
+    return 'unexpected-process';
+  }
+
+  try {
+    process.kill(pid, 'SIGTERM');
+  } catch {
+    return 'failed';
+  }
+
+  await new Promise((resolve) => setTimeout(resolve, 1000));
+
+  if (isProcessRunning(pid)) {
+    try {
+      process.kill(pid, 'SIGKILL');
+    } catch {
+      // ignore
+    }
+    await new Promise((resolve) => setTimeout(resolve, 500));
+    if (isProcessRunning(pid)) {
+      return 'failed';
+    }
+  }
+
+  try {
+    fs.unlinkSync(pidPath);
+  } catch {
+    // ignore
+  }
+
+  return 'stopped';
+}
+
+export const stopCommand: CommandModule = {
+  command: 'stop',
+  describe: 'Stop the LiteRT-LM server',
+  builder: (yargs) =>
+    yargs.option('port', {
+      type: 'number',
+      description: 'Port where the LiteRT server is running',
+    }),
+  handler: async (argv) => {
+    let port: number | undefined;
+    if (argv['port'] !== undefined) {
+      port = Number(argv['port']);
+    }
+
+    if (!port) {
+      const { configuredPort } = resolveGemmaConfig(DEFAULT_PORT);
+      port = configuredPort;
+    }
+
+    const processInfo = readServerProcessInfo();
+    const pid = processInfo?.pid ?? readServerPid();
+
+    if (pid !== null && isProcessRunning(pid)) {
+      debugLogger.log(`Stopping LiteRT server (PID ${pid})...`);
+      const result = await stopServer(port);
+      if (result === 'stopped') {
+        debugLogger.log(chalk.green('LiteRT server stopped.'));
+        await exitCli(0);
+      } else if (result === 'unexpected-process') {
+        debugLogger.error(
+          chalk.red(
+            `Refusing to stop PID ${pid} because it does not match the expected LiteRT server process.`,
+          ),
+        );
+        debugLogger.error(
+          chalk.dim(
+            'Remove the stale pid file after verifying the process, or stop the process manually.',
+          ),
+        );
+        await exitCli(1);
+      } else {
+        debugLogger.error(chalk.red('Failed to stop LiteRT server.'));
+        await exitCli(1);
+      }
+      return;
+    }
+
+    const running = await isServerRunning(port);
+    if (running) {
+      debugLogger.log(
+        chalk.yellow(
+          `A server is responding on port ${port}, but it was not started by "gemini gemma start".`,
+        ),
+      );
+      debugLogger.log(
+        chalk.dim(
+          'If you started it manually, stop it from the terminal where it is running.',
+        ),
+      );
+      await exitCli(1);
+    } else {
+      debugLogger.log('No LiteRT server is currently running.');
+      await exitCli(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();
     },
   );
 
@@ -338,6 +369,7 @@ describe('parseArguments', () => {
       { cmd: 'skill list', expected: true },
       { cmd: 'hooks migrate', expected: true },
       { cmd: 'hook migrate', expected: true },
+      { cmd: 'gemma status', expected: true },
       { cmd: 'some query', expected: undefined },
       { cmd: 'hello world', expected: undefined },
     ])(
@@ -561,7 +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');
       });
 
@@ -578,9 +610,6 @@ describe('parseArguments', () => {
           'Cannot use both --yolo (-y) and --approval-mode together. Use --approval-mode=yolo instead.',
         ),
       );
-
-      mockExit.mockRestore();
-      mockConsoleError.mockRestore();
     },
   );
 
@@ -605,7 +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');
     });
 
@@ -624,10 +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 () => {
@@ -758,6 +783,12 @@ describe('parseArguments', () => {
     const argv = await parseArguments(settings);
     expect(argv.isCommand).toBe(true);
   });
+
+  it('should set isCommand to true for gemma command', async () => {
+    process.argv = ['node', 'script.js', 'gemma', 'status'];
+    const argv = await parseArguments(createTestMergedSettings());
+    expect(argv.isCommand).toBe(true);
+  });
 });
 
 describe('loadCliConfig', () => {
@@ -773,6 +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 = [
@@ -865,16 +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(
@@ -888,8 +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 () => {
@@ -2210,51 +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);
   });
 });
@@ -3030,6 +3106,8 @@ describe('loadCliConfig gemmaModelRouter', () => {
       experimental: {
         gemmaModelRouter: {
           enabled: true,
+          autoStartServer: false,
+          binaryPath: '/custom/lit',
           classifier: {
             host: 'http://custom:1234',
             model: 'custom-gemma',
@@ -3040,10 +3118,24 @@ describe('loadCliConfig gemmaModelRouter', () => {
     const config = await loadCliConfig(settings, 'test-session', argv);
     expect(config.getGemmaModelRouterEnabled()).toBe(true);
     const gemmaSettings = config.getGemmaModelRouterSettings();
+    expect(gemmaSettings.autoStartServer).toBe(false);
+    expect(gemmaSettings.binaryPath).toBe('/custom/lit');
     expect(gemmaSettings.classifier?.host).toBe('http://custom:1234');
     expect(gemmaSettings.classifier?.model).toBe('custom-gemma');
   });
 
+  it('should load experimental.gemma setting from merged settings', async () => {
+    process.argv = ['node', 'script.js'];
+    const argv = await parseArguments(createTestMergedSettings());
+    const settings = createTestMergedSettings({
+      experimental: {
+        gemma: true,
+      },
+    });
+    const config = await loadCliConfig(settings, 'test-session', argv);
+    expect(config.getExperimentalGemma()).toBe(true);
+  });
+
   it('should handle partial gemmaModelRouter settings', async () => {
     process.argv = ['node', 'script.js'];
     const argv = await parseArguments(createTestMergedSettings());
@@ -3057,6 +3149,8 @@ describe('loadCliConfig gemmaModelRouter', () => {
     const config = await loadCliConfig(settings, 'test-session', argv);
     expect(config.getGemmaModelRouterEnabled()).toBe(true);
     const gemmaSettings = config.getGemmaModelRouterSettings();
+    expect(gemmaSettings.autoStartServer).toBe(false);
+    expect(gemmaSettings.binaryPath).toBe('');
     expect(gemmaSettings.classifier?.host).toBe('http://localhost:9379');
     expect(gemmaSettings.classifier?.model).toBe('gemma3-1b-gpu-custom');
   });
@@ -3200,7 +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');
     });
 
@@ -3218,10 +3312,6 @@ describe('Output format', () => {
       expect.stringContaining('Invalid values:'),
     );
     expect(mockConsoleError).toHaveBeenCalled();
-
-    mockExit.mockRestore();
-    mockConsoleError.mockRestore();
-    debugErrorSpy.mockRestore();
   });
 });
 
@@ -3252,13 +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(() => {});
@@ -3272,10 +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

@@ -13,6 +13,7 @@ import { mcpCommand } from '../commands/mcp.js';
 import { extensionsCommand } from '../commands/extensions.js';
 import { skillsCommand } from '../commands/skills.js';
 import { hooksCommand } from '../commands/hooks.js';
+import { gemmaCommand } from '../commands/gemma.js';
 import {
   setGeminiMdFilename as setServerGeminiMdFilename,
   getCurrentGeminiMdFilename,
@@ -23,6 +24,7 @@ import {
   FileDiscoveryService,
   resolveTelemetrySettings,
   FatalConfigError,
+  getErrorMessage,
   getPty,
   debugLogger,
   loadServerHierarchicalMemory,
@@ -46,7 +48,6 @@ import {
   type HookEventName,
   type OutputFormat,
   detectIdeFromEnv,
-  generalistProfile,
 } from '@google/gemini-cli-core';
 import {
   type Settings,
@@ -59,6 +60,7 @@ import {
 
 import { loadSandboxConfig } from './sandboxConfig.js';
 import { resolvePath } from '../utils/resolvePath.js';
+import { isRecord } from '../utils/settingsUtils.js';
 import { RESUME_LATEST } from '../utils/sessionUtils.js';
 
 import { isWorkspaceTrusted } from './trustedFolders.js';
@@ -94,6 +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;
@@ -105,6 +108,7 @@ export interface CliArgs {
   startupMessages?: string[];
   rawOutput: boolean | undefined;
   acceptRawOutputRisk: boolean | undefined;
+  skipTrust: boolean | undefined;
   isCommand: boolean | undefined;
 }
 
@@ -181,6 +185,7 @@ export async function parseArguments(
         extensionsCommand,
         skillsCommand,
         hooksCommand,
+        gemmaCommand,
       ];
 
       const subcommands = commandModules.flatMap((mod) => {
@@ -233,6 +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';
       }
@@ -260,6 +269,7 @@ export async function parseArguments(
   yargsInstance.command(extensionsCommand);
   yargsInstance.command(skillsCommand);
   yargsInstance.command(hooksCommand);
+  yargsInstance.command(gemmaCommand);
 
   yargsInstance
     .command('$0 [query..]', 'Launch Gemini CLI', (yargsInstance) =>
@@ -288,6 +298,11 @@ export async function parseArguments(
           description:
             'Execute the provided prompt and continue in interactive mode',
         })
+        .option('skip-trust', {
+          type: 'boolean',
+          description: 'Trust the current workspace for this session.',
+          default: false,
+        })
         .option('worktree', {
           alias: 'w',
           type: 'string',
@@ -396,6 +411,25 @@ export async function parseArguments(
             return trimmed;
           },
         })
+        .option('session-id', {
+          type: 'string',
+          nargs: 1,
+          description: 'Start a new session with a manually provided UUID.',
+          coerce: (value: string): string => {
+            const trimmed = value.trim();
+            if (!trimmed) {
+              throw new Error('The --session-id option cannot be empty.');
+            }
+            if (!/^[a-zA-Z0-9-_]+$/.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:
@@ -456,9 +490,16 @@ export async function parseArguments(
   yargsInstance.wrap(yargsInstance.terminalWidth());
   let result;
   try {
-    result = await yargsInstance.parse();
+    const parsed = await yargsInstance.parse();
+    if (!isRecord(parsed)) {
+      throw new Error('Failed to parse arguments');
+    }
+    result = parsed;
+    if (result['skip-trust']) {
+      process.env['GEMINI_CLI_TRUST_WORKSPACE'] = 'true';
+    }
   } catch (e) {
-    const msg = e instanceof Error ? e.message : String(e);
+    const msg = getErrorMessage(e);
     debugLogger.error(msg);
     yargsInstance.showHelp();
     await runExitCleanup();
@@ -472,11 +513,13 @@ export async function parseArguments(
   }
 
   // Normalize query args: handle both quoted "@path file" and unquoted @path file
-  // eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
-  const queryArg = (result as { query?: string | string[] | undefined }).query;
-  const q: string | undefined = Array.isArray(queryArg)
-    ? queryArg.join(' ')
-    : queryArg;
+  const queryArg = result['query'];
+  let q: string | undefined;
+  if (Array.isArray(queryArg)) {
+    q = queryArg.join(' ');
+  } else if (typeof queryArg === 'string') {
+    q = queryArg;
+  }
 
   // -p/--prompt forces non-interactive mode; positional args default to interactive in TTY
   if (q && !result['prompt']) {
@@ -491,8 +534,8 @@ export async function parseArguments(
   }
 
   // Keep CliArgs.query as a string for downstream typing
-  (result as Record<string, unknown>)['query'] = q || undefined;
-  (result as Record<string, unknown>)['startupMessages'] = startupMessages;
+  result['query'] = q || undefined;
+  result['startupMessages'] = startupMessages;
 
   // The import format is now only controlled by settings.memoryImportFormat
   // We no longer accept it as a CLI argument
@@ -544,7 +587,7 @@ export async function loadCliConfig(
       ? false
       : (settings.security?.folderTrust?.enabled ?? false);
   const trustedFolder =
-    isWorkspaceTrusted(settings, cwd, undefined, {
+    isWorkspaceTrusted(settings, cwd, {
       prompt: argv.prompt,
       query: argv.query,
     })?.isTrusted ?? false;
@@ -590,7 +633,7 @@ export async function loadCliConfig(
         return resolveToRealPath(trimmedPath) !== realCwd;
       } catch (e) {
         debugLogger.debug(
-          `[IDE] Skipping inaccessible workspace folder: ${trimmedPath} (${e instanceof Error ? e.message : String(e)})`,
+          `[IDE] Skipping inaccessible workspace folder: ${trimmedPath} (${getErrorMessage(e)})`,
         );
         return false;
       }
@@ -614,7 +657,7 @@ export async function loadCliConfig(
     .getExtensions()
     .find((ext) => ext.isActive && ext.plan?.directory)?.plan;
 
-  const experimentalJitContext = settings.experimental.jitContext;
+  const experimentalJitContext = settings.experimental.jitContext ?? true;
 
   let extensionRegistryURI =
     process.env['GEMINI_CLI_EXTENSION_REGISTRY_URI'] ??
@@ -798,9 +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
@@ -884,14 +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({
@@ -915,6 +970,7 @@ export async function loadCliConfig(
     worktreeSettings,
 
     coreTools: settings.tools?.core || undefined,
+    experimentalContextManagementConfig: profileSelector,
     allowedTools: allowedTools.length > 0 ? allowedTools : undefined,
     policyEngineConfig,
     policyUpdateConfirmationRequest,
@@ -980,6 +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
@@ -988,11 +1045,15 @@ export async function loadCliConfig(
     enableEventDrivenScheduler: true,
     skillsSupport: settings.skills?.enabled ?? true,
     disabledSkills: settings.skills?.disabled,
-    experimentalJitContext: settings.experimental?.jitContext,
-    experimentalMemoryManager: settings.experimental?.memoryManager,
+    experimentalJitContext,
+    experimentalMemoryV2: settings.experimental?.memoryV2,
+    experimentalAutoMemory: settings.experimental?.autoMemory,
+    experimentalGemma: settings.experimental?.gemma,
     contextManagement,
     modelSteering: settings.experimental?.modelSteering,
-    topicUpdateNarration: settings.experimental?.topicUpdateNarration,
+    topicUpdateNarration:
+      settings.general?.topicUpdateNarration ??
+      settings.experimental?.topicUpdateNarration,
     noBrowser: !!process.env['NO_BROWSER'],
     summarizeToolOutput: settings.model?.summarizeToolOutput,
     ideMode,
@@ -1026,6 +1087,7 @@ export async function loadCliConfig(
     recordResponses: argv.recordResponses,
     retryFetchErrors: settings.general?.retryFetchErrors,
     billing: settings.billing,
+    vertexAiRouting: settings.billing?.vertexAi,
     maxAttempts: settings.general?.maxAttempts,
     ptyInfo: ptyInfo?.name,
     disableLLMCorrection: settings.tools?.disableLLMCorrection,
@@ -1092,7 +1154,7 @@ async function resolveWorktreeSettings(
     worktreeBaseSha = stdout.trim();
   } catch (e: unknown) {
     debugLogger.debug(
-      `Failed to resolve worktree base SHA at ${worktreePath}: ${e instanceof Error ? e.message : String(e)}`,
+      `Failed to resolve worktree base SHA at ${worktreePath}: ${getErrorMessage(e)}`,
     );
   }
 

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

@@ -26,6 +26,7 @@ import {
   loadAgentsFromDirectory,
   loadSkillsFromDir,
   getRealPath,
+  normalizePath,
 } from '@google/gemini-cli-core';
 import {
   loadSettings,
@@ -1420,6 +1421,7 @@ name = "yolo-checker"
         '.gemini',
         'trustedFolders.json',
       );
+      vi.stubEnv('GEMINI_CLI_TRUSTED_FOLDERS_PATH', trustedFoldersPath);
       vi.mocked(isWorkspaceTrusted).mockReturnValue({
         isTrusted: false,
         source: undefined,
@@ -1438,7 +1440,9 @@ name = "yolo-checker"
       const trustedFolders = JSON.parse(
         fs.readFileSync(trustedFoldersPath, 'utf-8'),
       );
-      expect(trustedFolders[tempWorkspaceDir]).toBe('TRUST_FOLDER');
+      expect(trustedFolders[normalizePath(tempWorkspaceDir)]).toBe(
+        'TRUST_FOLDER',
+      );
     });
 
     describe.each([true, false])(

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 = {
@@ -1912,6 +2043,9 @@ describe('Settings Loading and Merging', () => {
       const geminiEnvPath = path.resolve(
         path.join(MOCK_WORKSPACE_DIR, GEMINI_DIR, '.env'),
       );
+      const workspaceEnvPath = path.resolve(
+        path.join(MOCK_WORKSPACE_DIR, '.env'),
+      );
 
       vi.spyOn(trustedFolders, 'isWorkspaceTrusted').mockReturnValue({
         isTrusted: isWorkspaceTrustedValue,
@@ -1919,9 +2053,11 @@ describe('Settings Loading and Merging', () => {
       });
       (mockFsExistsSync as Mock).mockImplementation((p: fs.PathLike) => {
         const normalizedP = path.resolve(p.toString());
-        return [path.resolve(USER_SETTINGS_PATH), geminiEnvPath].includes(
-          normalizedP,
-        );
+        return [
+          path.resolve(USER_SETTINGS_PATH),
+          geminiEnvPath,
+          workspaceEnvPath,
+        ].includes(normalizedP);
       });
       const userSettingsContent: Settings = {
         ui: {
@@ -1941,7 +2077,7 @@ describe('Settings Loading and Merging', () => {
           const normalizedP = path.resolve(p.toString());
           if (normalizedP === path.resolve(USER_SETTINGS_PATH))
             return JSON.stringify(userSettingsContent);
-          if (normalizedP === geminiEnvPath)
+          if (normalizedP === geminiEnvPath || normalizedP === workspaceEnvPath)
             return 'TESTTEST=1234\nGEMINI_API_KEY=test-key';
           return '{}';
         },
@@ -1970,7 +2106,7 @@ describe('Settings Loading and Merging', () => {
       expect(process.env['TESTTEST']).not.toEqual('1234');
     });
 
-    it('does load env files from untrusted spaces when NOT sandboxed', () => {
+    it('does NOT load non-whitelisted env files from untrusted spaces even when NOT sandboxed', () => {
       setup({ isFolderTrustEnabled: true, isWorkspaceTrustedValue: false });
       const settings = {
         security: { folderTrust: { enabled: true } },
@@ -1978,7 +2114,8 @@ describe('Settings Loading and Merging', () => {
       } as Settings;
       loadEnvironment(settings, MOCK_WORKSPACE_DIR, isWorkspaceTrusted);
 
-      expect(process.env['TESTTEST']).toEqual('1234');
+      expect(process.env['TESTTEST']).not.toEqual('1234');
+      expect(process.env['GEMINI_API_KEY']).toEqual('test-key');
     });
 
     it('does not load env files when trust is undefined and sandboxed', () => {

packages/cli/src/config/settings.ts

@@ -78,7 +78,12 @@ export function getMergeStrategyForPath(
 
 export const USER_SETTINGS_PATH = Storage.getGlobalSettingsPath();
 export const USER_SETTINGS_DIR = path.dirname(USER_SETTINGS_PATH);
-export const DEFAULT_EXCLUDED_ENV_VARS = ['DEBUG', 'DEBUG_MODE'];
+export const DEFAULT_EXCLUDED_ENV_VARS = [
+  'DEBUG',
+  'DEBUG_MODE',
+  'GEMINI_CLI_IDE_SERVER_STDIO_COMMAND',
+  'GEMINI_CLI_IDE_SERVER_STDIO_ARGS',
+];
 
 const AUTH_ENV_VAR_WHITELIST = [
   'GEMINI_API_KEY',
@@ -494,13 +499,15 @@ export class LoadedSettings {
   }
 }
 
-function findEnvFile(startDir: string): string | null {
+function findEnvFile(startDir: string, isTrusted: boolean): string | null {
   let currentDir = path.resolve(startDir);
   while (true) {
     // prefer gemini-specific .env under GEMINI_DIR
-    const geminiEnvPath = path.join(currentDir, GEMINI_DIR, '.env');
-    if (fs.existsSync(geminiEnvPath)) {
-      return geminiEnvPath;
+    if (isTrusted) {
+      const geminiEnvPath = path.join(currentDir, GEMINI_DIR, '.env');
+      if (fs.existsSync(geminiEnvPath)) {
+        return geminiEnvPath;
+      }
     }
     const envPath = path.join(currentDir, '.env');
     if (fs.existsSync(envPath)) {
@@ -509,9 +516,11 @@ function findEnvFile(startDir: string): string | null {
     const parentDir = path.dirname(currentDir);
     if (parentDir === currentDir || !parentDir) {
       // check .env under home as fallback, again preferring gemini-specific .env
-      const homeGeminiEnvPath = path.join(homedir(), GEMINI_DIR, '.env');
-      if (fs.existsSync(homeGeminiEnvPath)) {
-        return homeGeminiEnvPath;
+      if (isTrusted) {
+        const homeGeminiEnvPath = path.join(homedir(), GEMINI_DIR, '.env');
+        if (fs.existsSync(homeGeminiEnvPath)) {
+          return homeGeminiEnvPath;
+        }
       }
       const homeEnvPath = path.join(homedir(), '.env');
       if (fs.existsSync(homeEnvPath)) {
@@ -554,10 +563,10 @@ export function loadEnvironment(
   workspaceDir: string,
   isWorkspaceTrustedFn = isWorkspaceTrusted,
 ): void {
-  const envFilePath = findEnvFile(workspaceDir);
   const trustResult = isWorkspaceTrustedFn(settings, workspaceDir);
-
   const isTrusted = trustResult.isTrusted ?? false;
+  const envFilePath = findEnvFile(workspaceDir, isTrusted);
+
   // Check settings OR check process.argv directly since this might be called
   // before arguments are fully parsed. This is a best-effort sniffing approach
   // that happens early in the CLI lifecycle. It is designed to detect the
@@ -592,8 +601,8 @@ export function loadEnvironment(
       for (const key in parsedEnv) {
         if (Object.hasOwn(parsedEnv, key)) {
           let value = parsedEnv[key];
-          // If the workspace is untrusted but we are sandboxed, only allow whitelisted variables.
-          if (!isTrusted && isSandboxed) {
+          // If the workspace is untrusted, only allow whitelisted variables.
+          if (!isTrusted) {
             if (!AUTH_ENV_VAR_WHITELIST.includes(key)) {
               continue;
             }
@@ -664,7 +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');
@@ -680,14 +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,
@@ -698,9 +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({
@@ -709,33 +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.test.ts

@@ -138,6 +138,10 @@ describe('SettingsSchema', () => {
         getSettingsSchema().context.properties.fileFiltering.properties
           ?.enableRecursiveFileSearch,
       ).toBeDefined();
+      expect(
+        getSettingsSchema().context.properties.fileFiltering.properties
+          ?.enableFileWatcher,
+      ).toBeDefined();
       expect(
         getSettingsSchema().context.properties.fileFiltering.properties
           ?.customIgnoreFilePaths,
@@ -313,6 +317,22 @@ describe('SettingsSchema', () => {
       ).toBe(false);
     });
 
+    it('should have Vertex AI routing settings in schema', () => {
+      const vertexAi =
+        getSettingsSchema().billing.properties.vertexAi.properties;
+
+      expect(vertexAi.requestType).toBeDefined();
+      expect(vertexAi.requestType.type).toBe('enum');
+      expect(
+        vertexAi.requestType.options?.map((option) => option.value),
+      ).toEqual(['dedicated', 'shared']);
+      expect(vertexAi.sharedRequestType).toBeDefined();
+      expect(vertexAi.sharedRequestType.type).toBe('enum');
+      expect(
+        vertexAi.sharedRequestType.options?.map((option) => option.value),
+      ).toEqual(['priority', 'flex']);
+    });
+
     it('should have folderTrustFeature setting in schema', () => {
       expect(
         getSettingsSchema().security.properties.folderTrust.properties.enabled,
@@ -471,11 +491,33 @@ describe('SettingsSchema', () => {
       expect(enabled.category).toBe('Experimental');
       expect(enabled.default).toBe(false);
       expect(enabled.requiresRestart).toBe(true);
-      expect(enabled.showInDialog).toBe(false);
+      expect(enabled.showInDialog).toBe(true);
       expect(enabled.description).toBe(
         'Enable the Gemma Model Router (experimental). Requires a local endpoint serving Gemma via the Gemini API using LiteRT-LM shim.',
       );
 
+      const autoStartServer = gemmaModelRouter.properties.autoStartServer;
+      expect(autoStartServer).toBeDefined();
+      expect(autoStartServer.type).toBe('boolean');
+      expect(autoStartServer.category).toBe('Experimental');
+      expect(autoStartServer.default).toBe(false);
+      expect(autoStartServer.requiresRestart).toBe(true);
+      expect(autoStartServer.showInDialog).toBe(true);
+      expect(autoStartServer.description).toBe(
+        'Automatically start the LiteRT-LM server when Gemini CLI starts and the Gemma router is enabled.',
+      );
+
+      const binaryPath = gemmaModelRouter.properties.binaryPath;
+      expect(binaryPath).toBeDefined();
+      expect(binaryPath.type).toBe('string');
+      expect(binaryPath.category).toBe('Experimental');
+      expect(binaryPath.default).toBe('');
+      expect(binaryPath.requiresRestart).toBe(true);
+      expect(binaryPath.showInDialog).toBe(false);
+      expect(binaryPath.description).toBe(
+        'Custom path to the LiteRT-LM binary. Leave empty to use the default location (~/.gemini/bin/litert/).',
+      );
+
       const classifier = gemmaModelRouter.properties.classifier;
       expect(classifier).toBeDefined();
       expect(classifier.type).toBe('object');

packages/cli/src/config/settingsSchema.ts

@@ -21,6 +21,7 @@ import {
   type AgentOverride,
   type CustomTheme,
   type SandboxConfig,
+  type VertexAiRoutingConfig,
 } from '@google/gemini-cli-core';
 import type { SessionRetentionSettings } from './settings.js';
 import { DEFAULT_MIN_RETENTION } from '../utils/sessionCleanup.js';
@@ -256,14 +257,29 @@ const SETTINGS_SCHEMA = {
       },
       enableNotifications: {
         type: 'boolean',
-        label: 'Enable Notifications',
+        label: 'Enable Terminal Notifications',
         category: 'General',
         requiresRestart: false,
         default: false,
         description:
-          'Enable run-event notifications for action-required prompts and session completion.',
+          'Enable terminal run-event notifications for action-required prompts and session completion.',
         showInDialog: true,
       },
+      notificationMethod: {
+        type: 'enum',
+        label: 'Terminal Notification Method',
+        category: 'General',
+        requiresRestart: false,
+        default: 'auto',
+        description: 'How to send terminal notifications.',
+        showInDialog: true,
+        options: [
+          { value: 'auto', label: 'Auto' },
+          { value: 'osc9', label: 'OSC 9' },
+          { value: 'osc777', label: 'OSC 777' },
+          { value: 'bell', label: 'Bell' },
+        ],
+      },
       checkpointing: {
         type: 'object',
         label: 'Checkpointing',
@@ -403,6 +419,16 @@ const SETTINGS_SCHEMA = {
         },
         description: 'Settings for automatic session cleanup.',
       },
+      topicUpdateNarration: {
+        type: 'boolean',
+        label: 'Topic & Update Narration',
+        category: 'General',
+        requiresRestart: false,
+        default: true,
+        description:
+          'Enable the Topic & Update communication model for reduced chattiness and structured progress reporting.',
+        showInDialog: true,
+      },
     },
   },
   output: {
@@ -965,6 +991,45 @@ const SETTINGS_SCHEMA = {
           { value: 'never', label: 'Never use credits' },
         ],
       },
+      vertexAi: {
+        type: 'object',
+        label: 'Vertex AI',
+        category: 'Advanced',
+        requiresRestart: true,
+        default: undefined as VertexAiRoutingConfig | undefined,
+        description: 'Vertex AI request routing settings.',
+        showInDialog: false,
+        properties: {
+          requestType: {
+            type: 'enum',
+            label: 'Vertex AI Request Type',
+            category: 'Advanced',
+            requiresRestart: true,
+            default: undefined as VertexAiRoutingConfig['requestType'],
+            description:
+              'Sets the X-Vertex-AI-LLM-Request-Type header for Vertex AI requests.',
+            showInDialog: false,
+            options: [
+              { value: 'dedicated', label: 'Dedicated' },
+              { value: 'shared', label: 'Shared' },
+            ],
+          },
+          sharedRequestType: {
+            type: 'enum',
+            label: 'Vertex AI Shared Request Type',
+            category: 'Advanced',
+            requiresRestart: true,
+            default: undefined as VertexAiRoutingConfig['sharedRequestType'],
+            description:
+              'Sets the X-Vertex-AI-LLM-Shared-Request-Type header for Vertex AI requests.',
+            showInDialog: false,
+            options: [
+              { value: 'priority', label: 'Priority' },
+              { value: 'flex', label: 'Flex' },
+            ],
+          },
+        },
+      },
     },
   },
 
@@ -1406,6 +1471,17 @@ const SETTINGS_SCHEMA = {
             description: 'Respect .geminiignore files when searching.',
             showInDialog: true,
           },
+          enableFileWatcher: {
+            type: 'boolean',
+            label: 'Enable File Watcher',
+            category: 'Context',
+            requiresRestart: true,
+            default: false,
+            description: oneLine`
+              Enable file watcher updates for @ file suggestions (experimental).
+            `,
+            showInDialog: false,
+          },
           enableRecursiveFileSearch: {
             type: 'boolean',
             label: 'Enable Recursive File Search',
@@ -1591,6 +1667,19 @@ const SETTINGS_SCHEMA = {
         showInDialog: false,
         items: { type: 'string' },
       },
+      confirmationRequired: {
+        type: 'array',
+        label: 'Confirmation Required',
+        category: 'Advanced',
+        requiresRestart: true,
+        default: undefined as string[] | undefined,
+        description: oneLine`
+          Tool names that always require user confirmation.
+          Takes precedence over allowed tools and core tool allowlists.
+        `,
+        showInDialog: false,
+        items: { type: 'string' },
+      },
       exclude: {
         type: 'array',
         label: 'Exclude Tools',
@@ -1963,6 +2052,96 @@ const SETTINGS_SCHEMA = {
     description: 'Setting to enable experimental features',
     showInDialog: false,
     properties: {
+      gemma: {
+        type: 'boolean',
+        label: 'Gemma Models',
+        category: 'Experimental',
+        requiresRestart: true,
+        default: false,
+        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',
@@ -2064,8 +2243,9 @@ const SETTINGS_SCHEMA = {
         label: 'JIT Context Loading',
         category: 'Experimental',
         requiresRestart: true,
-        default: false,
-        description: 'Enable Just-In-Time (JIT) context loading.',
+        default: true,
+        description:
+          'Enable Just-In-Time (JIT) context loading. Defaults to true; set to false to opt out and load all GEMINI.md files into the system instruction up-front.',
         showInDialog: false,
       },
       useOSC52Paste: {
@@ -2144,6 +2324,26 @@ const SETTINGS_SCHEMA = {
             default: false,
             description:
               'Enable the Gemma Model Router (experimental). Requires a local endpoint serving Gemma via the Gemini API using LiteRT-LM shim.',
+            showInDialog: true,
+          },
+          autoStartServer: {
+            type: 'boolean',
+            label: 'Auto-start LiteRT Server',
+            category: 'Experimental',
+            requiresRestart: true,
+            default: false,
+            description:
+              'Automatically start the LiteRT-LM server when Gemini CLI starts and the Gemma router is enabled.',
+            showInDialog: true,
+          },
+          binaryPath: {
+            type: 'string',
+            label: 'LiteRT Binary Path',
+            category: 'Experimental',
+            requiresRestart: true,
+            default: '',
+            description:
+              'Custom path to the LiteRT-LM binary. Leave empty to use the default location (~/.gemini/bin/litert/).',
             showInDialog: false,
           },
           classifier: {
@@ -2178,14 +2378,35 @@ const SETTINGS_SCHEMA = {
           },
         },
       },
-      memoryManager: {
+      memoryV2: {
         type: 'boolean',
-        label: 'Memory Manager Agent',
+        label: 'Memory v2',
+        category: 'Experimental',
+        requiresRestart: true,
+        default: true,
+        description:
+          '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:
-          'Replace the built-in save_memory tool with a memory manager subagent that supports adding, removing, de-duplicating, and organizing memories.',
+          'Significantly lowers token limits to force early garbage collection and distillation for testing purposes.',
+        showInDialog: false,
+      },
+      autoMemory: {
+        type: 'boolean',
+        label: 'Auto Memory',
+        category: 'Experimental',
+        requiresRestart: true,
+        default: false,
+        description:
+          'Automatically extract reusable skills from past sessions in the background. Review results with /memory inbox.',
         showInDialog: true,
       },
       generalistProfile: {
@@ -2213,9 +2434,8 @@ const SETTINGS_SCHEMA = {
         category: 'Experimental',
         requiresRestart: false,
         default: false,
-        description:
-          'Enable the experimental Topic & Update communication model for reduced chattiness and structured progress reporting.',
-        showInDialog: true,
+        description: 'Deprecated: Use general.topicUpdateNarration instead.',
+        showInDialog: false,
       },
     },
   },
@@ -2966,6 +3186,11 @@ export const SETTINGS_SCHEMA_DEFINITIONS: Record<
         description: 'Protocol for OTLP exporters.',
         enum: ['grpc', 'http'],
       },
+      traces: {
+        type: 'boolean',
+        description:
+          'Whether detailed traces with large attributes are captured.',
+      },
       logPrompts: {
         type: 'boolean',
         description: 'Whether prompts are logged in telemetry payloads.',
@@ -3064,6 +3289,7 @@ export const SETTINGS_SCHEMA_DEFINITIONS: Record<
           secondary: { type: 'string' },
           link: { type: 'string' },
           accent: { type: 'string' },
+          response: { type: 'string' },
         },
       },
       background: {

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

@@ -11,7 +11,7 @@ import * as os from 'node:os';
 import {
   FatalConfigError,
   ideContextStore,
-  coreEvents,
+  normalizePath,
 } from '@google/gemini-cli-core';
 import {
   loadTrustedFolders,
@@ -32,9 +32,14 @@ vi.mock('@google/gemini-cli-core', async (importOriginal) => {
     ...actual,
     homedir: () => '/mock/home/user',
     isHeadlessMode: vi.fn(() => false),
-    coreEvents: {
-      emitFeedback: vi.fn(),
-    },
+    coreEvents: Object.assign(
+      Object.create(Object.getPrototypeOf(actual.coreEvents)),
+      actual.coreEvents,
+      {
+        emitFeedback: vi.fn(),
+      },
+    ),
+    FatalConfigError: actual.FatalConfigError,
   };
 });
 
@@ -53,6 +58,7 @@ describe('Trusted Folders', () => {
     // Reset the internal state
     resetTrustedFoldersForTesting();
     vi.clearAllMocks();
+    delete process.env['GEMINI_CLI_TRUST_WORKSPACE'];
   });
 
   afterEach(() => {
@@ -70,8 +76,14 @@ describe('Trusted Folders', () => {
 
       // Start two concurrent calls
       // These will race to acquire the lock on the real file system
-      const p1 = loadedFolders.setValue('/path1', TrustLevel.TRUST_FOLDER);
-      const p2 = loadedFolders.setValue('/path2', TrustLevel.TRUST_FOLDER);
+      const p1 = loadedFolders.setValue(
+        path.resolve('/path1'),
+        TrustLevel.TRUST_FOLDER,
+      );
+      const p2 = loadedFolders.setValue(
+        path.resolve('/path2'),
+        TrustLevel.TRUST_FOLDER,
+      );
 
       await Promise.all([p1, p2]);
 
@@ -80,8 +92,8 @@ describe('Trusted Folders', () => {
       const config = JSON.parse(content);
 
       expect(config).toEqual({
-        '/path1': TrustLevel.TRUST_FOLDER,
-        '/path2': TrustLevel.TRUST_FOLDER,
+        [normalizePath('/path1')]: TrustLevel.TRUST_FOLDER,
+        [normalizePath('/path2')]: TrustLevel.TRUST_FOLDER,
       });
     });
   });
@@ -95,13 +107,16 @@ describe('Trusted Folders', () => {
 
     it('should load rules from the configuration file', () => {
       const config = {
-        '/user/folder': TrustLevel.TRUST_FOLDER,
+        [normalizePath('/user/folder')]: TrustLevel.TRUST_FOLDER,
       };
       fs.writeFileSync(trustedFoldersPath, JSON.stringify(config), 'utf-8');
 
       const { rules, errors } = loadTrustedFolders();
       expect(rules).toEqual([
-        { path: '/user/folder', trustLevel: TrustLevel.TRUST_FOLDER },
+        {
+          path: normalizePath('/user/folder'),
+          trustLevel: TrustLevel.TRUST_FOLDER,
+        },
       ]);
       expect(errors).toEqual([]);
     });
@@ -143,14 +158,14 @@ describe('Trusted Folders', () => {
       const content = `
         {
           // This is a comment
-          "/path": "TRUST_FOLDER"
+          "${normalizePath('/path').replaceAll('\\', '\\\\')}": "TRUST_FOLDER"
         }
       `;
       fs.writeFileSync(trustedFoldersPath, content, 'utf-8');
 
       const { rules, errors } = loadTrustedFolders();
       expect(rules).toEqual([
-        { path: '/path', trustLevel: TrustLevel.TRUST_FOLDER },
+        { path: normalizePath('/path'), trustLevel: TrustLevel.TRUST_FOLDER },
       ]);
       expect(errors).toEqual([]);
     });
@@ -216,15 +231,18 @@ describe('Trusted Folders', () => {
       fs.writeFileSync(trustedFoldersPath, '{}', 'utf-8');
       const loadedFolders = loadTrustedFolders();
 
-      await loadedFolders.setValue('/new/path', TrustLevel.TRUST_FOLDER);
+      await loadedFolders.setValue(
+        normalizePath('/new/path'),
+        TrustLevel.TRUST_FOLDER,
+      );
 
-      expect(loadedFolders.user.config['/new/path']).toBe(
+      expect(loadedFolders.user.config[normalizePath('/new/path')]).toBe(
         TrustLevel.TRUST_FOLDER,
       );
 
       const content = fs.readFileSync(trustedFoldersPath, 'utf-8');
       const config = JSON.parse(content);
-      expect(config['/new/path']).toBe(TrustLevel.TRUST_FOLDER);
+      expect(config[normalizePath('/new/path')]).toBe(TrustLevel.TRUST_FOLDER);
     });
 
     it('should throw FatalConfigError if there were load errors', async () => {
@@ -237,28 +255,6 @@ describe('Trusted Folders', () => {
         loadedFolders.setValue('/some/path', TrustLevel.TRUST_FOLDER),
       ).rejects.toThrow(FatalConfigError);
     });
-
-    it('should report corrupted config via coreEvents.emitFeedback and still succeed', async () => {
-      // Initialize with valid JSON
-      fs.writeFileSync(trustedFoldersPath, '{}', 'utf-8');
-      const loadedFolders = loadTrustedFolders();
-
-      // Corrupt the file after initial load
-      fs.writeFileSync(trustedFoldersPath, 'invalid json', 'utf-8');
-
-      await loadedFolders.setValue('/new/path', TrustLevel.TRUST_FOLDER);
-
-      expect(coreEvents.emitFeedback).toHaveBeenCalledWith(
-        'error',
-        expect.stringContaining('may be corrupted'),
-        expect.any(Error),
-      );
-
-      // Should have overwritten the corrupted file with new valid config
-      const content = fs.readFileSync(trustedFoldersPath, 'utf-8');
-      const config = JSON.parse(content);
-      expect(config).toEqual({ '/new/path': TrustLevel.TRUST_FOLDER });
-    });
   });
 
   describe('isWorkspaceTrusted Integration', () => {
@@ -427,16 +423,28 @@ describe('Trusted Folders', () => {
       },
     };
 
-    it('should return true when isHeadlessMode is true, ignoring config', async () => {
+    it('should NOT return true when isHeadlessMode is true, ignoring config', async () => {
       const geminiCore = await import('@google/gemini-cli-core');
       vi.spyOn(geminiCore, 'isHeadlessMode').mockReturnValue(true);
 
       expect(isWorkspaceTrusted(mockSettings)).toEqual({
-        isTrusted: true,
+        isTrusted: undefined,
         source: undefined,
       });
     });
 
+    it('should return true when GEMINI_CLI_TRUST_WORKSPACE is true', async () => {
+      process.env['GEMINI_CLI_TRUST_WORKSPACE'] = 'true';
+      try {
+        expect(isWorkspaceTrusted(mockSettings)).toEqual({
+          isTrusted: true,
+          source: 'env',
+        });
+      } finally {
+        delete process.env['GEMINI_CLI_TRUST_WORKSPACE'];
+      }
+    });
+
     it('should fall back to config when isHeadlessMode is false', async () => {
       const geminiCore = await import('@google/gemini-cli-core');
       vi.spyOn(geminiCore, 'isHeadlessMode').mockReturnValue(false);
@@ -449,12 +457,12 @@ describe('Trusted Folders', () => {
       );
     });
 
-    it('should return true for isPathTrusted when isHeadlessMode is true', async () => {
+    it('should return undefined for isPathTrusted when isHeadlessMode is true', async () => {
       const geminiCore = await import('@google/gemini-cli-core');
       vi.spyOn(geminiCore, 'isHeadlessMode').mockReturnValue(true);
 
       const folders = loadTrustedFolders();
-      expect(folders.isPathTrusted('/any-untrusted-path')).toBe(true);
+      expect(folders.isPathTrusted('/any-untrusted-path')).toBe(undefined);
     });
   });