Compare commits

...

2 Commits

Author SHA1 Message Date
jacob314 1817ce8bad Polish UX to be width agnostic 2026-03-04 00:43:08 -08:00
jacob314 2717cacb8c test(cli): fix flaky QuotaDisplay snapshot and env leakage in StatusDisplay 2026-03-02 12:22:11 -08:00
118 changed files with 4673 additions and 894 deletions
+1 -1
View File
@@ -22,7 +22,7 @@ get_issue_labels() {
# Check cache
case "${ISSUE_LABELS_CACHE_FLAT}" in
*"|${ISSUE_NUM}:"*)
local suffix="${ISSUE_LABELS_CACHE_FLAT#*|${ISSUE_NUM}:}"
local suffix="${ISSUE_LABELS_CACHE_FLAT#*|"${ISSUE_NUM}":}"
echo "${suffix%%|*}"
return
;;
+2 -2
View File
@@ -224,8 +224,6 @@ jobs:
if: |
always() && (needs.merge_queue_skipper.result !='success' || needs.merge_queue_skipper.outputs.skip != 'true')
runs-on: 'gemini-cli-windows-16-core'
continue-on-error: true
steps:
- name: 'Checkout'
uses: 'actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955' # ratchet:actions/checkout@v5
@@ -315,6 +313,7 @@ jobs:
needs:
- 'e2e_linux'
- 'e2e_mac'
- 'e2e_windows'
- 'evals'
- 'merge_queue_skipper'
runs-on: 'gemini-cli-ubuntu-16-core'
@@ -323,6 +322,7 @@ jobs:
run: |
if [[ ${{ needs.e2e_linux.result }} != 'success' || \
${{ needs.e2e_mac.result }} != 'success' || \
${{ needs.e2e_windows.result }} != 'success' || \
${{ needs.evals.result }} != 'success' ]]; then
echo "One or more E2E jobs failed."
exit 1
+2 -1
View File
@@ -360,7 +360,6 @@ jobs:
runs-on: 'gemini-cli-windows-16-core'
needs: 'merge_queue_skipper'
if: "${{needs.merge_queue_skipper.outputs.skip == 'false'}}"
continue-on-error: true
timeout-minutes: 60
strategy:
matrix:
@@ -458,6 +457,7 @@ jobs:
- 'link_checker'
- 'test_linux'
- 'test_mac'
- 'test_windows'
- 'codeql'
- 'bundle_size'
runs-on: 'gemini-cli-ubuntu-16-core'
@@ -468,6 +468,7 @@ jobs:
(${{ needs.link_checker.result }} != 'success' && ${{ needs.link_checker.result }} != 'skipped') || \
(${{ needs.test_linux.result }} != 'success' && ${{ needs.test_linux.result }} != 'skipped') || \
(${{ needs.test_mac.result }} != 'success' && ${{ needs.test_mac.result }} != 'skipped') || \
(${{ needs.test_windows.result }} != 'success' && ${{ needs.test_windows.result }} != 'skipped') || \
(${{ needs.codeql.result }} != 'success' && ${{ needs.codeql.result }} != 'skipped') || \
(${{ needs.bundle_size.result }} != 'success' && ${{ needs.bundle_size.result }} != 'skipped') ]]; then
echo "One or more CI jobs failed."
+1
View File
@@ -27,6 +27,7 @@ jobs:
fail-fast: false
matrix:
model:
- 'gemini-3.1-pro-preview-customtools'
- 'gemini-3-pro-preview'
- 'gemini-3-flash-preview'
- 'gemini-2.5-pro'
+29
View File
@@ -0,0 +1,29 @@
# yaml-language-server: $schema=https://json.schemastore.org/github-workflow.json
name: 'PR rate limiter'
permissions: {}
on:
pull_request_target:
types:
- 'opened'
- 'reopened'
jobs:
limit:
runs-on: 'gemini-cli-ubuntu-16-core'
permissions:
contents: 'read'
pull-requests: 'write'
steps:
- name: 'Limit open pull requests per user'
uses: 'Homebrew/actions/limit-pull-requests@9ceb7934560eb61d131dde205a6c2d77b2e1529d' # master
with:
except-author-associations: 'MEMBER,OWNER,COLLABORATOR'
comment-limit: 8
comment: >
You already have 7 pull requests open. Please work on getting
existing PRs merged before opening more.
close-limit: 8
close: true
+10 -8
View File
@@ -29,6 +29,7 @@ they appear in the UI.
| Enable Auto Update | `general.enableAutoUpdate` | Enable automatic updates. | `true` |
| Enable Notifications | `general.enableNotifications` | Enable run-event notifications for action-required prompts and session completion. Currently macOS only. | `false` |
| Plan Directory | `general.plan.directory` | The directory where planning artifacts are stored. If not specified, defaults to the system temporary directory. | `undefined` |
| 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 | `false` |
| Keep chat history | `general.sessionRetention.maxAge` | Automatically delete chats older than this time period (e.g., "30d", "7d", "24h", "1w") | `undefined` |
@@ -111,14 +112,15 @@ they appear in the UI.
### Security
| UI Label | Setting | Description | Default |
| ------------------------------------- | ----------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------- |
| Disable YOLO Mode | `security.disableYoloMode` | Disable YOLO mode, even if enabled by a flag. | `false` |
| Allow Permanent Tool Approval | `security.enablePermanentToolApproval` | Enable the "Allow for all future sessions" option in tool confirmation dialogs. | `false` |
| Blocks extensions from Git | `security.blockGitExtensions` | Blocks installing and loading extensions from Git. | `false` |
| Extension Source Regex Allowlist | `security.allowedExtensions` | List of Regex patterns for allowed extensions. If nonempty, only extensions that match the patterns in this list are allowed. Overrides the blockGitExtensions setting. | `[]` |
| Folder Trust | `security.folderTrust.enabled` | Setting to track whether Folder trust is enabled. | `true` |
| Enable Environment Variable Redaction | `security.environmentVariableRedaction.enabled` | Enable redaction of environment variables that may contain secrets. | `false` |
| UI Label | Setting | Description | Default |
| ------------------------------------- | ----------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ------- |
| Disable YOLO Mode | `security.disableYoloMode` | Disable YOLO mode, even if enabled by a flag. | `false` |
| Allow Permanent Tool Approval | `security.enablePermanentToolApproval` | Enable the "Allow for all future sessions" option in tool confirmation dialogs. | `false` |
| Blocks extensions from Git | `security.blockGitExtensions` | Blocks installing and loading extensions from Git. | `false` |
| Extension Source Regex Allowlist | `security.allowedExtensions` | List of Regex patterns for allowed extensions. If nonempty, only extensions that match the patterns in this list are allowed. Overrides the blockGitExtensions setting. | `[]` |
| Folder Trust | `security.folderTrust.enabled` | Setting to track whether Folder trust is enabled. | `true` |
| Enable Environment Variable Redaction | `security.environmentVariableRedaction.enabled` | Enable redaction of environment variables that may contain secrets. | `false` |
| Enable Context-Aware Security | `security.enableConseca` | Enable the context-aware security checker. This feature uses an LLM to dynamically generate and enforce security policies for tool use based on your prompt, providing an additional layer of protection against unintended actions. | `false` |
### Advanced
+38 -14
View File
@@ -116,7 +116,9 @@ The manifest file defines the extension's behavior and configuration.
"description": "My awesome extension",
"mcpServers": {
"my-server": {
"command": "node my-server.js"
"command": "node",
"args": ["${extensionPath}/my-server.js"],
"cwd": "${extensionPath}"
}
},
"contextFileName": "GEMINI.md",
@@ -124,19 +126,41 @@ The manifest file defines the extension's behavior and configuration.
}
```
- `name`: A unique identifier for the extension. Use lowercase letters, numbers,
and dashes. This name must match the extension's directory name.
- `version`: The current version of the extension.
- `description`: A short summary shown in the extension gallery.
- <a id="mcp-servers"></a>`mcpServers`: A map of Model Context Protocol (MCP)
servers. Extension servers follow the same format as standard
[CLI configuration](../reference/configuration.md).
- `contextFileName`: The name of the context file (defaults to `GEMINI.md`). Can
also be an array of strings to load multiple context files.
- `excludeTools`: An array of tools to block from the model. You can restrict
specific arguments, such as `run_shell_command(rm -rf)`.
- `themes`: An optional list of themes provided by the extension. See
[Themes](../cli/themes.md) for more information.
- `name`: The name of the extension. This is used to uniquely identify the
extension and for conflict resolution when extension commands have the same
name as user or project commands. The name should be lowercase or numbers and
use dashes instead of underscores or spaces. This is how users will refer to
your extension in the CLI. Note that we expect this name to match the
extension directory name.
- `version`: The version of the extension.
- `description`: A short description of the extension. This will be displayed on
[geminicli.com/extensions](https://geminicli.com/extensions).
- `mcpServers`: A map of MCP servers to settings. The key is the name of the
server, and the value is the server configuration. These servers will be
loaded on startup just like MCP servers defined in a
[`settings.json` file](../reference/configuration.md). If both an extension
and a `settings.json` file define an MCP server with the same name, the server
defined in the `settings.json` file takes precedence.
- Note that all MCP server configuration options are supported except for
`trust`.
- For portability, you should use `${extensionPath}` to refer to files within
your extension directory.
- Separate your executable and its arguments using `command` and `args`
instead of putting them both in `command`.
- `contextFileName`: The name of the file that contains the context for the
extension. This will be used to load the context from the extension directory.
If this property is not used but a `GEMINI.md` file is present in your
extension directory, then that file will be loaded.
- `excludeTools`: An array of tool names to exclude from the model. You can also
specify command-specific restrictions for tools that support it, like the
`run_shell_command` tool. For example,
`"excludeTools": ["run_shell_command(rm -rf)"]` will block the `rm -rf`
command. Note that this differs from the MCP server `excludeTools`
functionality, which can be listed in the MCP server config.
When Gemini CLI starts, it loads all the extensions and merges their
configurations. If there are any conflicts, the workspace configuration takes
precedence.
### Extension settings
+8
View File
@@ -98,6 +98,8 @@ and parameter rewriting.
- `tool_name`: (`string`) The name of the tool being called.
- `tool_input`: (`object`) The raw arguments generated by the model.
- `mcp_context`: (`object`) Optional metadata for MCP-based tools.
- `original_request_name`: (`string`) The original name of the tool being
called, if this is a tail tool call.
- **Relevant Output Fields**:
- `decision`: Set to `"deny"` (or `"block"`) to prevent the tool from
executing.
@@ -120,12 +122,18 @@ hiding sensitive output from the agent.
- `tool_response`: (`object`) The result containing `llmContent`,
`returnDisplay`, and optional `error`.
- `mcp_context`: (`object`)
- `original_request_name`: (`string`) The original name of the tool being
called, if this is a tail tool call.
- **Relevant Output Fields**:
- `decision`: Set to `"deny"` to hide the real tool output from the agent.
- `reason`: Required if denied. This text **replaces** the tool result sent
back to the model.
- `hookSpecificOutput.additionalContext`: Text that is **appended** to the
tool result for the agent.
- `hookSpecificOutput.tailToolCallRequest`: (`{ name: string, args: object }`)
A request to execute another tool immediately after this one. The result of
this "tail call" will replace the original tool's response. Ideal for
programmatic tool routing.
- `continue`: Set to `false` to **kill the entire agent loop** immediately.
- **Exit Code 2 (Block Result)**: Hides the tool result. Uses `stderr` as the
replacement content sent to the agent. **The turn continues.**
+13
View File
@@ -142,6 +142,11 @@ their corresponding top-level category object in your `settings.json` file.
request" errors.
- **Default:** `false`
- **`general.maxAttempts`** (number):
- **Description:** Maximum number of attempts for requests to the main chat
model. Cannot exceed 10.
- **Default:** `10`
- **`general.debugKeystrokeLogging`** (boolean):
- **Description:** Enable debug logging of keystrokes to the console.
- **Default:** `false`
@@ -868,6 +873,14 @@ their corresponding top-level category object in your `settings.json` file.
- **Default:** `undefined`
- **Requires restart:** Yes
- **`security.enableConseca`** (boolean):
- **Description:** Enable the context-aware security checker. This feature
uses an LLM to dynamically generate and enforce security policies for tool
use based on your prompt, providing an additional layer of protection
against unintended actions.
- **Default:** `false`
- **Requires restart:** Yes
#### `advanced`
- **`advanced.autoConfigureMemory`** (boolean):
+62 -2
View File
@@ -163,7 +163,8 @@ Each server configuration supports the following properties:
- **`args`** (string[]): Command-line arguments for Stdio transport
- **`headers`** (object): Custom HTTP headers when using `url` or `httpUrl`
- **`env`** (object): Environment variables for the server process. Values can
reference environment variables using `$VAR_NAME` or `${VAR_NAME}` syntax
reference environment variables using `$VAR_NAME` or `${VAR_NAME}` syntax (all
platforms), or `%VAR_NAME%` (Windows only).
- **`cwd`** (string): Working directory for Stdio transport
- **`timeout`** (number): Request timeout in milliseconds (default: 600,000ms =
10 minutes)
@@ -184,6 +185,63 @@ Each server configuration supports the following properties:
Service Account to impersonate. Used with
`authProviderType: 'service_account_impersonation'`.
### Environment variable expansion
Gemini CLI automatically expands environment variables in the `env` block of
your MCP server configuration. This allows you to securely reference variables
defined in your shell or environment without hardcoding sensitive information
directly in your `settings.json` file.
The expansion utility supports:
- **POSIX/Bash syntax:** `$VARIABLE_NAME` or `${VARIABLE_NAME}` (supported on
all platforms)
- **Windows syntax:** `%VARIABLE_NAME%` (supported only when running on Windows)
If a variable is not defined in the current environment, it resolves to an empty
string.
**Example:**
```json
"env": {
"API_KEY": "$MY_EXTERNAL_TOKEN",
"LOG_LEVEL": "$LOG_LEVEL",
"TEMP_DIR": "%TEMP%"
}
```
### Security and environment sanitization
To protect your credentials, Gemini CLI performs environment sanitization when
spawning MCP server processes.
#### Automatic redaction
By default, the CLI redacts sensitive environment variables from the base
environment (inherited from the host process) to prevent unintended exposure to
third-party MCP servers. This includes:
- Core project keys: `GEMINI_API_KEY`, `GOOGLE_API_KEY`, etc.
- Variables matching sensitive patterns: `*TOKEN*`, `*SECRET*`, `*PASSWORD*`,
`*KEY*`, `*AUTH*`, `*CREDENTIAL*`.
- Certificates and private key patterns.
#### Explicit overrides
If an environment variable must be passed to an MCP server, you must explicitly
state it in the `env` property of the server configuration in `settings.json`.
Explicitly defined variables (including those from extensions) are trusted and
are **not** subjected to the automatic redaction process.
This follows the security principle that if a variable is explicitly configured
by the user for a specific server, it constitutes informed consent to share that
specific data with that server.
> **Note:** Even when explicitly defined, you should avoid hardcoding secrets.
> Instead, use environment variable expansion (e.g., `"MY_KEY": "$MY_KEY"`) to
> securely pull the value from your host environment at runtime.
### OAuth support for remote MCP servers
The Gemini CLI supports OAuth 2.0 authentication for remote MCP servers using
@@ -738,7 +796,9 @@ The MCP integration tracks several states:
- **Trust settings:** The `trust` option bypasses all confirmation dialogs. Use
cautiously and only for servers you completely control
- **Access tokens:** Be security-aware when configuring environment variables
containing API keys or tokens
containing API keys or tokens. See
[Security and environment sanitization](#security-and-environment-sanitization)
for details on how Gemini CLI protects your credentials.
- **Sandbox compatibility:** When using sandboxing, ensure MCP servers are
available within the sandbox environment
- **Private data:** Using broadly scoped personal access tokens can lead to
+13 -12
View File
@@ -78,22 +78,23 @@ describe('Frugal reads eval', () => {
).toBe(true);
let totalLinesRead = 0;
const readRanges: { offset: number; limit: number }[] = [];
const readRanges: { start_line: number; end_line: number }[] = [];
for (const call of targetFileReads) {
const args = JSON.parse(call.toolRequest.args);
expect(
args.limit,
'Agent read the entire file (missing limit) instead of using ranged read',
args.end_line,
'Agent read the entire file (missing end_line) instead of using ranged read',
).toBeDefined();
const limit = args.limit;
const offset = args.offset ?? 0;
totalLinesRead += limit;
readRanges.push({ offset, limit });
const end_line = args.end_line;
const start_line = args.start_line ?? 1;
const linesRead = end_line - start_line + 1;
totalLinesRead += linesRead;
readRanges.push({ start_line, end_line });
expect(args.limit, 'Agent read too many lines at once').toBeLessThan(
expect(linesRead, 'Agent read too many lines at once').toBeLessThan(
1001,
);
}
@@ -108,7 +109,7 @@ describe('Frugal reads eval', () => {
const errorLines = [500, 510, 520];
for (const line of errorLines) {
const covered = readRanges.some(
(range) => line >= range.offset && line < range.offset + range.limit,
(range) => line >= range.start_line && line <= range.end_line,
);
expect(covered, `Agent should have read around line ${line}`).toBe(
true,
@@ -191,8 +192,8 @@ describe('Frugal reads eval', () => {
for (const call of targetFileReads) {
const args = JSON.parse(call.toolRequest.args);
expect(
args.limit,
'Agent should have used ranged read (limit) to save tokens',
args.end_line,
'Agent should have used ranged read (end_line) to save tokens',
).toBeDefined();
}
},
@@ -253,7 +254,7 @@ describe('Frugal reads eval', () => {
// and just read the whole file to be efficient with tool calls.
const readEntireFile = targetFileReads.some((call) => {
const args = JSON.parse(call.toolRequest.args);
return args.limit === undefined;
return args.end_line === undefined;
});
expect(
+2 -2
View File
@@ -68,7 +68,7 @@ describe('Frugal Search', () => {
const args = getParams(call);
return (
args.file_path === 'src/legacy_processor.ts' &&
(args.limit === undefined || args.limit === null)
(args.end_line === undefined || args.end_line === null)
);
});
@@ -87,7 +87,7 @@ describe('Frugal Search', () => {
if (
call.toolRequest.name === 'read_file' &&
args.file_path === 'src/legacy_processor.ts' &&
args.limit !== undefined
args.end_line !== undefined
) {
return true;
}
+1 -1
View File
@@ -56,7 +56,7 @@ describe('interactive_commands', () => {
const scaffoldCall = logs.find(
(l) =>
l.toolRequest.name === 'run_shell_command' &&
/npm (init|create)|npx create-|yarn create|pnpm create/.test(
/npm (init|create)|npx (.*)?create-|yarn create|pnpm create/.test(
l.toolRequest.args,
),
);
@@ -0,0 +1,2 @@
{"method":"generateContentStream","response":[{"candidates":[{"content":{"parts":[{"functionCall":{"name":"read_file","args":{"file_path":"original.txt"}}}],"role":"model"},"finishReason":"STOP","index":0}]}]}
{"method":"generateContentStream","response":[{"candidates":[{"content":{"parts":[{"text":"Tail call completed successfully."}],"role":"model"},"finishReason":"STOP","index":0}]}]}
+107
View File
@@ -286,6 +286,113 @@ describe('Hooks System Integration', () => {
});
});
describe('Command Hooks - Tail Tool Calls', () => {
it('should execute a tail tool call from AfterTool hooks and replace original response', async () => {
// Create a script that acts as the hook.
// It will trigger on "read_file" and issue a tail call to "write_file".
rig.setup('should execute a tail tool call from AfterTool hooks', {
fakeResponsesPath: join(
import.meta.dirname,
'hooks-system.tail-tool-call.responses',
),
});
const hookOutput = {
decision: 'allow',
hookSpecificOutput: {
hookEventName: 'AfterTool',
tailToolCallRequest: {
name: 'write_file',
args: {
file_path: 'tail-called-file.txt',
content: 'Content from tail call',
},
},
},
};
const hookScript = `console.log(JSON.stringify(${JSON.stringify(
hookOutput,
)})); process.exit(0);`;
const scriptPath = join(rig.testDir!, 'tail_call_hook.js');
writeFileSync(scriptPath, hookScript);
const commandPath = scriptPath.replace(/\\/g, '/');
rig.setup('should execute a tail tool call from AfterTool hooks', {
fakeResponsesPath: join(
import.meta.dirname,
'hooks-system.tail-tool-call.responses',
),
settings: {
hooksConfig: {
enabled: true,
},
hooks: {
AfterTool: [
{
matcher: 'read_file',
hooks: [
{
type: 'command',
command: `node "${commandPath}"`,
timeout: 5000,
},
],
},
],
},
},
});
// Create a test file to trigger the read_file tool
rig.createFile('original.txt', 'Original content');
const cliOutput = await rig.run({
args: 'Read original.txt', // Fake responses should trigger read_file on this
});
// 1. Verify that write_file was called (as a tail call replacing read_file)
// Since read_file was replaced before finalizing, it will not appear in the tool logs.
const foundWriteFile = await rig.waitForToolCall('write_file');
expect(foundWriteFile).toBeTruthy();
// Ensure hook logs are flushed and the final LLM response is received.
// The mock LLM is configured to respond with "Tail call completed successfully."
expect(cliOutput).toContain('Tail call completed successfully.');
// Ensure telemetry is written to disk
await rig.waitForTelemetryReady();
// Read hook logs to debug
const hookLogs = rig.readHookLogs();
const relevantHookLog = hookLogs.find(
(l) => l.hookCall.hook_event_name === 'AfterTool',
);
expect(relevantHookLog).toBeDefined();
// 2. Verify write_file was executed.
// In non-interactive mode, the CLI deduplicates tool execution logs by callId.
// Since a tail call reuses the original callId, "Tool: write_file" is not printed.
// Instead, we verify the side-effect (file creation) and the telemetry log.
// 3. Verify the tail-called tool actually wrote the file
const modifiedContent = rig.readFile('tail-called-file.txt');
expect(modifiedContent).toBe('Content from tail call');
// 4. Verify telemetry for the final tool call.
// The original 'read_file' call is replaced, so only 'write_file' is finalized and logged.
const toolLogs = rig.readToolLogs();
const successfulTools = toolLogs.filter((t) => t.toolRequest.success);
expect(
successfulTools.some((t) => t.toolRequest.name === 'write_file'),
).toBeTruthy();
// The original request name should be preserved in the log payload if possible,
// but the executed tool name is 'write_file'.
});
});
describe('BeforeModel Hooks - LLM Request Modification', () => {
it('should modify LLM requests with BeforeModel hooks', async () => {
// Create a hook script that replaces the LLM request with a modified version
+382 -428
View File
File diff suppressed because it is too large Load Diff
+9 -10
View File
@@ -29,6 +29,8 @@ import {
CoderAgentEvent,
getPersistedState,
setPersistedState,
getContextIdFromMetadata,
getAgentSettingsFromMetadata,
} from '../types.js';
import { loadConfig, loadEnvironment, setTargetDir } from '../config/config.js';
import { loadSettings } from '../config/settings.js';
@@ -117,8 +119,7 @@ export class CoderAgentExecutor implements AgentExecutor {
const agentSettings = persistedState._agentSettings;
const config = await this.getConfig(agentSettings, sdkTask.id);
const contextId: string =
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
(metadata['_contextId'] as string) || sdkTask.contextId;
getContextIdFromMetadata(metadata) || sdkTask.contextId;
const runtimeTask = await Task.create(
sdkTask.id,
contextId,
@@ -141,8 +142,10 @@ export class CoderAgentExecutor implements AgentExecutor {
agentSettingsInput?: AgentSettings,
eventBus?: ExecutionEventBus,
): Promise<TaskWrapper> {
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
const agentSettings = agentSettingsInput || ({} as AgentSettings);
const agentSettings: AgentSettings = agentSettingsInput || {
kind: CoderAgentEvent.StateAgentSettingsEvent,
workspacePath: process.cwd(),
};
const config = await this.getConfig(agentSettings, taskId);
const runtimeTask = await Task.create(
taskId,
@@ -292,8 +295,7 @@ export class CoderAgentExecutor implements AgentExecutor {
const contextId: string =
userMessage.contextId ||
sdkTask?.contextId ||
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
(sdkTask?.metadata?.['_contextId'] as string) ||
getContextIdFromMetadata(sdkTask?.metadata) ||
uuidv4();
logger.info(
@@ -388,10 +390,7 @@ export class CoderAgentExecutor implements AgentExecutor {
}
} else {
logger.info(`[CoderAgentExecutor] Creating new task ${taskId}.`);
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
const agentSettings = userMessage.metadata?.[
'coderAgent'
] as AgentSettings;
const agentSettings = getAgentSettingsFromMetadata(userMessage.metadata);
try {
wrapper = await this.createTask(
taskId,
+8 -2
View File
@@ -513,7 +513,10 @@ describe('Task', () => {
{
request: { callId: '1' },
status: 'awaiting_approval',
confirmationDetails: { onConfirm: onConfirmSpy },
confirmationDetails: {
type: 'edit',
onConfirm: onConfirmSpy,
},
},
] as unknown as ToolCall[];
@@ -533,7 +536,10 @@ describe('Task', () => {
{
request: { callId: '1' },
status: 'awaiting_approval',
confirmationDetails: { onConfirm: onConfirmSpy },
confirmationDetails: {
type: 'edit',
onConfirm: onConfirmSpy,
},
},
] as unknown as ToolCall[];
+99 -41
View File
@@ -59,6 +59,33 @@ import type { PartUnion, Part as genAiPart } from '@google/genai';
type UnionKeys<T> = T extends T ? keyof T : never;
type ConfirmationType = ToolCallConfirmationDetails['type'];
const VALID_CONFIRMATION_TYPES: readonly ConfirmationType[] = [
'edit',
'exec',
'mcp',
'info',
'ask_user',
'exit_plan_mode',
] as const;
function isToolCallConfirmationDetails(
value: unknown,
): value is ToolCallConfirmationDetails {
if (
typeof value !== 'object' ||
value === null ||
!('onConfirm' in value) ||
typeof value.onConfirm !== 'function' ||
!('type' in value) ||
typeof value.type !== 'string'
) {
return false;
}
return (VALID_CONFIRMATION_TYPES as readonly string[]).includes(value.type);
}
export class Task {
id: string;
contextId: string;
@@ -376,11 +403,10 @@ export class Task {
}
if (tc.status === 'awaiting_approval' && tc.confirmationDetails) {
this.pendingToolConfirmationDetails.set(
tc.request.callId,
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
tc.confirmationDetails as ToolCallConfirmationDetails,
);
const details = tc.confirmationDetails;
if (isToolCallConfirmationDetails(details)) {
this.pendingToolConfirmationDetails.set(tc.request.callId, details);
}
}
// Only send an update if the status has actually changed.
@@ -412,11 +438,12 @@ export class Task {
);
toolCalls.forEach((tc: ToolCall) => {
if (tc.status === 'awaiting_approval' && tc.confirmationDetails) {
// eslint-disable-next-line @typescript-eslint/no-floating-promises, @typescript-eslint/no-unsafe-type-assertion
(tc.confirmationDetails as ToolCallConfirmationDetails).onConfirm(
ToolConfirmationOutcome.ProceedOnce,
);
this.pendingToolConfirmationDetails.delete(tc.request.callId);
const details = tc.confirmationDetails;
if (isToolCallConfirmationDetails(details)) {
// eslint-disable-next-line @typescript-eslint/no-floating-promises
details.onConfirm(ToolConfirmationOutcome.ProceedOnce);
this.pendingToolConfirmationDetails.delete(tc.request.callId);
}
}
});
return;
@@ -466,15 +493,13 @@ export class Task {
T extends ToolCall | AnyDeclarativeTool,
K extends UnionKeys<T>,
>(from: T, ...fields: K[]): Partial<T> {
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
const ret = {} as Pick<T, K>;
const ret: Partial<T> = {};
for (const field of fields) {
if (field in from) {
if (field in from && from[field] !== undefined) {
ret[field] = from[field];
}
}
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
return ret as Partial<T>;
return ret;
}
private toolStatusMessage(
@@ -485,8 +510,11 @@ export class Task {
const messageParts: Part[] = [];
// Create a serializable version of the ToolCall (pick necessary
// properties/avoid methods causing circular reference errors)
const serializableToolCall: Partial<ToolCall> = this._pickFields(
// properties/avoid methods causing circular reference errors).
// Type allows tool to be Partial<AnyDeclarativeTool> for serialization.
const serializableToolCall: Partial<Omit<ToolCall, 'tool'>> & {
tool?: Partial<AnyDeclarativeTool>;
} = this._pickFields(
tc,
'request',
'status',
@@ -496,8 +524,7 @@ export class Task {
);
if (tc.tool) {
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
serializableToolCall.tool = this._pickFields(
const toolFields = this._pickFields(
tc.tool,
'name',
'displayName',
@@ -507,7 +534,8 @@ export class Task {
'canUpdateOutput',
'schema',
'parameterSchema',
) as AnyDeclarativeTool;
);
serializableToolCall.tool = toolFields;
}
messageParts.push({
@@ -530,8 +558,15 @@ export class Task {
old_string: string,
new_string: string,
): Promise<string> {
// Validate path to prevent path traversal vulnerabilities
const resolvedPath = path.resolve(this.config.getTargetDir(), file_path);
const pathError = this.config.validatePathAccess(resolvedPath, 'read');
if (pathError) {
throw new Error(`Path validation failed: ${pathError}`);
}
try {
const currentContent = await fs.readFile(file_path, 'utf8');
const currentContent = await fs.readFile(resolvedPath, 'utf8');
return this._applyReplacement(
currentContent,
old_string,
@@ -625,15 +660,32 @@ export class Task {
request.args['old_string'] &&
request.args['new_string']
) {
const newContent = await this.getProposedContent(
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
request.args['file_path'] as string,
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
request.args['old_string'] as string,
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
request.args['new_string'] as string,
);
return { ...request, args: { ...request.args, newContent } };
const filePath = request.args['file_path'];
const oldString = request.args['old_string'];
const newString = request.args['new_string'];
if (
typeof filePath === 'string' &&
typeof oldString === 'string' &&
typeof newString === 'string'
) {
// Resolve and validate path to prevent path traversal (user-controlled file_path).
const resolvedPath = path.resolve(
this.config.getTargetDir(),
filePath,
);
const pathError = this.config.validatePathAccess(
resolvedPath,
'read',
);
if (!pathError) {
const newContent = await this.getProposedContent(
resolvedPath,
oldString,
newString,
);
return { ...request, args: { ...request.args, newContent } };
}
}
}
return request;
}),
@@ -725,10 +777,17 @@ export class Task {
break;
case GeminiEventType.Error:
default: {
// Block scope for lexical declaration
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
const errorEvent = event as ServerGeminiErrorEvent; // Type assertion
const errorMessage = errorEvent.value?.error
// Use type guard instead of unsafe type assertion
let errorEvent: ServerGeminiErrorEvent | undefined;
if (
event.type === GeminiEventType.Error &&
event.value &&
typeof event.value === 'object' &&
'error' in event.value
) {
errorEvent = event;
}
const errorMessage = errorEvent?.value?.error
? getErrorMessage(errorEvent.value.error)
: 'Unknown error from LLM stream';
logger.error(
@@ -737,7 +796,7 @@ export class Task {
);
let errMessage = `Unknown error from LLM stream: ${JSON.stringify(event)}`;
if (errorEvent.value?.error) {
if (errorEvent?.value?.error) {
errMessage = parseAndFormatApiError(errorEvent.value.error);
}
this.cancelPendingTools(`LLM stream error: ${errorMessage}`);
@@ -814,12 +873,11 @@ export class Task {
// If `edit` tool call, pass updated payload if presesent
if (confirmationDetails.type === 'edit') {
const payload = part.data['newContent']
? ({
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
newContent: part.data['newContent'] as string,
} as ToolConfirmationPayload)
: undefined;
const newContent = part.data['newContent'];
const payload =
typeof newContent === 'string'
? ({ newContent } as ToolConfirmationPayload)
: undefined;
this.skipFinalTrueAfterInlineEdit = !!payload;
try {
await confirmationDetails.onConfirm(confirmationOutcome, payload);
+51 -2
View File
@@ -122,11 +122,60 @@ export type PersistedTaskMetadata = { [k: string]: unknown };
export const METADATA_KEY = '__persistedState';
function isAgentSettings(value: unknown): value is AgentSettings {
return (
typeof value === 'object' &&
value !== null &&
'kind' in value &&
value.kind === CoderAgentEvent.StateAgentSettingsEvent &&
'workspacePath' in value &&
typeof value.workspacePath === 'string'
);
}
function isPersistedStateMetadata(
value: unknown,
): value is PersistedStateMetadata {
return (
typeof value === 'object' &&
value !== null &&
'_agentSettings' in value &&
'_taskState' in value &&
isAgentSettings(value._agentSettings)
);
}
export function getPersistedState(
metadata: PersistedTaskMetadata,
): PersistedStateMetadata | undefined {
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
return metadata?.[METADATA_KEY] as PersistedStateMetadata | undefined;
const state = metadata?.[METADATA_KEY];
if (isPersistedStateMetadata(state)) {
return state;
}
return undefined;
}
export function getContextIdFromMetadata(
metadata: PersistedTaskMetadata | undefined,
): string | undefined {
if (!metadata) {
return undefined;
}
const contextId = metadata['_contextId'];
return typeof contextId === 'string' ? contextId : undefined;
}
export function getAgentSettingsFromMetadata(
metadata: PersistedTaskMetadata | undefined,
): AgentSettings | undefined {
if (!metadata) {
return undefined;
}
const coderAgent = metadata['coderAgent'];
if (isAgentSettings(coderAgent)) {
return coderAgent;
}
return undefined;
}
export function setPersistedState(
@@ -71,6 +71,7 @@ export function createMockConfig(
getMcpServers: vi.fn().mockReturnValue({}),
}),
getGitService: vi.fn(),
validatePathAccess: vi.fn().mockReturnValue(undefined),
...overrides,
} as unknown as Config;
mockConfig.getMessageBus = vi.fn().mockReturnValue(createMockMessageBus());
+1
View File
@@ -878,6 +878,7 @@ export async function loadCliConfig(
agents: refreshedSettings.merged.agents,
};
},
enableConseca: settings.security?.enableConseca,
});
}
@@ -352,6 +352,38 @@ describe('Policy Engine Integration Tests', () => {
).toBe(PolicyDecision.DENY);
});
it('should correctly match tool annotations', async () => {
const settings: Settings = {};
const config = await createPolicyEngineConfig(
settings,
ApprovalMode.DEFAULT,
);
// Add a manual rule with annotations to the config
config.rules = config.rules || [];
config.rules.push({
toolAnnotations: { readOnlyHint: true },
decision: PolicyDecision.ALLOW,
priority: 10,
});
const engine = new PolicyEngine(config);
// A tool with readOnlyHint=true should be ALLOWED
const roCall = { name: 'some_tool', args: {} };
const roMeta = { readOnlyHint: true };
expect((await engine.check(roCall, undefined, roMeta)).decision).toBe(
PolicyDecision.ALLOW,
);
// A tool without the hint (or with false) should follow default decision (ASK_USER)
const rwMeta = { readOnlyHint: false };
expect((await engine.check(roCall, undefined, rwMeta)).decision).toBe(
PolicyDecision.ASK_USER,
);
});
describe.each(['write_file', 'replace'])(
'Plan Mode policy for %s',
(toolName) => {
+44
View File
@@ -142,4 +142,48 @@ describe('resolveWorkspacePolicyState', () => {
expect.stringContaining('Automatically accepting and loading'),
);
});
it('should not return workspace policies if cwd is the home directory', async () => {
const policiesDir = path.join(tempDir, '.gemini', 'policies');
fs.mkdirSync(policiesDir, { recursive: true });
fs.writeFileSync(path.join(policiesDir, 'policy.toml'), 'rules = []');
// Run from HOME directory (tempDir is mocked as HOME in beforeEach)
const result = await resolveWorkspacePolicyState({
cwd: tempDir,
trustedFolder: true,
interactive: true,
});
expect(result.workspacePoliciesDir).toBeUndefined();
expect(result.policyUpdateConfirmationRequest).toBeUndefined();
});
it('should not return workspace policies if cwd is a symlink to the home directory', async () => {
const policiesDir = path.join(tempDir, '.gemini', 'policies');
fs.mkdirSync(policiesDir, { recursive: true });
fs.writeFileSync(path.join(policiesDir, 'policy.toml'), 'rules = []');
// Create a symlink to the home directory
const symlinkDir = path.join(
os.tmpdir(),
`gemini-cli-symlink-${Date.now()}`,
);
fs.symlinkSync(tempDir, symlinkDir, 'dir');
try {
// Run from symlink to HOME directory
const result = await resolveWorkspacePolicyState({
cwd: symlinkDir,
trustedFolder: true,
interactive: true,
});
expect(result.workspacePoliciesDir).toBeUndefined();
expect(result.policyUpdateConfirmationRequest).toBeUndefined();
} finally {
// Clean up symlink
fs.unlinkSync(symlinkDir);
}
});
});
+9 -3
View File
@@ -67,9 +67,15 @@ export async function resolveWorkspacePolicyState(options: {
| undefined;
if (trustedFolder) {
const potentialWorkspacePoliciesDir = new Storage(
cwd,
).getWorkspacePoliciesDir();
const storage = new Storage(cwd);
// If we are in the home directory (or rather, our target Gemini dir is the global one),
// don't treat it as a workspace to avoid loading global policies twice.
if (storage.isWorkspaceHomeDir()) {
return { workspacePoliciesDir: undefined };
}
const potentialWorkspacePoliciesDir = storage.getWorkspacePoliciesDir();
const integrityManager = new PolicyIntegrityManager();
const integrityResult = await integrityManager.checkIntegrity(
'workspace',
+41 -7
View File
@@ -79,6 +79,7 @@ import {
import {
FatalConfigError,
GEMINI_DIR,
Storage,
type MCPServerConfig,
} from '@google/gemini-cli-core';
import { updateSettingsFilePreservingFormat } from '../utils/commentJson.js';
@@ -126,6 +127,30 @@ vi.mock('@google/gemini-cli-core', async (importOriginal) => {
const actual =
await importOriginal<typeof import('@google/gemini-cli-core')>();
const os = await import('node:os');
const pathMod = await import('node:path');
const fsMod = await import('node:fs');
// Helper to resolve paths using the test's mocked environment
const testResolve = (p: string | undefined) => {
if (!p) return '';
try {
// Use the mocked fs.realpathSync if available, otherwise fallback
return fsMod.realpathSync(pathMod.resolve(p));
} catch {
return pathMod.resolve(p);
}
};
// Create a smarter mock for isWorkspaceHomeDir
vi.spyOn(actual.Storage.prototype, 'isWorkspaceHomeDir').mockImplementation(
function (this: Storage) {
const target = testResolve(pathMod.dirname(this.getGeminiDir()));
// Pick up the mocked home directory specifically from the 'os' mock
const home = testResolve(os.homedir());
return actual.normalizePath(target) === actual.normalizePath(home);
},
);
return {
...actual,
coreEvents: mockCoreEvents,
@@ -1491,20 +1516,29 @@ describe('Settings Loading and Merging', () => {
return pStr;
});
// Force the storage check to return true for this specific test
const isWorkspaceHomeDirSpy = vi
.spyOn(Storage.prototype, 'isWorkspaceHomeDir')
.mockReturnValue(true);
(mockFsExistsSync as Mock).mockImplementation(
(p: string) =>
// Only return true for workspace settings path to see if it gets loaded
p === mockWorkspaceSettingsPath,
);
const settings = loadSettings(mockSymlinkDir);
try {
const settings = loadSettings(mockSymlinkDir);
// Verify that even though the file exists, it was NOT loaded because realpath matched home
expect(fs.readFileSync).not.toHaveBeenCalledWith(
mockWorkspaceSettingsPath,
'utf-8',
);
expect(settings.workspace.settings).toEqual({});
// Verify that even though the file exists, it was NOT loaded because realpath matched home
expect(fs.readFileSync).not.toHaveBeenCalledWith(
mockWorkspaceSettingsPath,
'utf-8',
);
expect(settings.workspace.settings).toEqual({});
} finally {
isWorkspaceHomeDirSpy.mockRestore();
}
});
});
+5 -21
View File
@@ -637,24 +637,8 @@ export function loadSettings(
const systemSettingsPath = getSystemSettingsPath();
const systemDefaultsPath = getSystemDefaultsPath();
// Resolve paths to their canonical representation to handle symlinks
const resolvedWorkspaceDir = path.resolve(workspaceDir);
const resolvedHomeDir = path.resolve(homedir());
let realWorkspaceDir = resolvedWorkspaceDir;
try {
// fs.realpathSync gets the "true" path, resolving any symlinks
realWorkspaceDir = fs.realpathSync(resolvedWorkspaceDir);
} catch (_e) {
// This is okay. The path might not exist yet, and that's a valid state.
}
// We expect homedir to always exist and be resolvable.
const realHomeDir = fs.realpathSync(resolvedHomeDir);
const workspaceSettingsPath = new Storage(
workspaceDir,
).getWorkspaceSettingsPath();
const storage = new Storage(workspaceDir);
const workspaceSettingsPath = storage.getWorkspaceSettingsPath();
const load = (filePath: string): { settings: Settings; rawJson?: string } => {
try {
@@ -712,7 +696,7 @@ export function loadSettings(
settings: {} as Settings,
rawJson: undefined,
};
if (realWorkspaceDir !== realHomeDir) {
if (!storage.isWorkspaceHomeDir()) {
workspaceResult = load(workspaceSettingsPath);
}
@@ -800,11 +784,11 @@ export function loadSettings(
readOnly: false,
},
{
path: realWorkspaceDir === realHomeDir ? '' : workspaceSettingsPath,
path: storage.isWorkspaceHomeDir() ? '' : workspaceSettingsPath,
settings: workspaceSettings,
originalSettings: workspaceOriginalSettings,
rawJson: workspaceResult.rawJson,
readOnly: realWorkspaceDir === realHomeDir,
readOnly: storage.isWorkspaceHomeDir(),
},
isTrusted,
settingsErrors,
+20
View File
@@ -297,6 +297,16 @@ const SETTINGS_SCHEMA = {
'Retry on "exception TypeError: fetch failed sending request" errors.',
showInDialog: false,
},
maxAttempts: {
type: 'number',
label: 'Max Chat Model Attempts',
category: 'General',
requiresRestart: false,
default: 10,
description:
'Maximum number of attempts for requests to the main chat model. Cannot exceed 10.',
showInDialog: true,
},
debugKeystrokeLogging: {
type: 'boolean',
label: 'Debug Keystroke Logging',
@@ -1483,6 +1493,16 @@ const SETTINGS_SCHEMA = {
},
},
},
enableConseca: {
type: 'boolean',
label: 'Enable Context-Aware Security',
category: 'Security',
requiresRestart: true,
default: false,
description:
'Enable the context-aware security checker. This feature uses an LLM to dynamically generate and enforce security policies for tool use based on your prompt, providing an additional layer of protection against unintended actions.',
showInDialog: true,
},
},
},
@@ -182,7 +182,7 @@ describe('<Footer />', () => {
},
);
await waitUntilReady();
expect(lastFrame()).toContain('15%');
expect(lastFrame()).toContain('85%');
expect(lastFrame()).toMatchSnapshot();
unmount();
});
@@ -5,10 +5,20 @@
*/
import { render } from '../../test-utils/render.js';
import { describe, it, expect } from 'vitest';
import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest';
import { QuotaDisplay } from './QuotaDisplay.js';
describe('QuotaDisplay', () => {
beforeEach(() => {
vi.stubEnv('TZ', 'America/Los_Angeles');
vi.useFakeTimers();
vi.setSystemTime(new Date('2026-03-02T20:29:00.000Z'));
});
afterEach(() => {
vi.useRealTimers();
vi.unstubAllEnvs();
});
it('should not render when remaining is undefined', async () => {
const { lastFrame, waitUntilReady, unmount } = render(
<QuotaDisplay remaining={undefined} limit={100} />,
@@ -36,7 +46,7 @@ describe('QuotaDisplay', () => {
unmount();
});
it('should not render when usage > 20%', async () => {
it('should not render when usage < 80%', async () => {
const { lastFrame, waitUntilReady, unmount } = render(
<QuotaDisplay remaining={85} limit={100} />,
);
@@ -45,7 +55,7 @@ describe('QuotaDisplay', () => {
unmount();
});
it('should render yellow when usage < 20%', async () => {
it('should render warning when used >= 80%', async () => {
const { lastFrame, waitUntilReady, unmount } = render(
<QuotaDisplay remaining={15} limit={100} />,
);
@@ -54,7 +64,7 @@ describe('QuotaDisplay', () => {
unmount();
});
it('should render red when usage < 5%', async () => {
it('should render critical when used >= 95%', async () => {
const { lastFrame, waitUntilReady, unmount } = render(
<QuotaDisplay remaining={4} limit={100} />,
);
+18 -16
View File
@@ -7,9 +7,9 @@
import type React from 'react';
import { Text } from 'ink';
import {
getStatusColor,
QUOTA_THRESHOLD_HIGH,
QUOTA_THRESHOLD_MEDIUM,
getUsedStatusColor,
QUOTA_USED_WARNING_THRESHOLD,
QUOTA_USED_CRITICAL_THRESHOLD,
} from '../utils/displayUtils.js';
import { formatResetTime } from '../utils/formatters.js';
@@ -30,26 +30,24 @@ export const QuotaDisplay: React.FC<QuotaDisplayProps> = ({
return null;
}
const percentage = (remaining / limit) * 100;
const usedPercentage = 100 - (remaining / limit) * 100;
if (percentage > QUOTA_THRESHOLD_HIGH) {
if (usedPercentage < QUOTA_USED_WARNING_THRESHOLD) {
return null;
}
const color = getStatusColor(percentage, {
green: QUOTA_THRESHOLD_HIGH,
yellow: QUOTA_THRESHOLD_MEDIUM,
const color = getUsedStatusColor(usedPercentage, {
warning: QUOTA_USED_WARNING_THRESHOLD,
critical: QUOTA_USED_CRITICAL_THRESHOLD,
});
const resetInfo =
!terse && resetTime ? `, ${formatResetTime(resetTime)}` : '';
if (remaining === 0) {
const resetMsg = resetTime
? `, resets in ${formatResetTime(resetTime, true)}`
: '';
return (
<Text color={color}>
{terse
? 'Limit reached'
: `/stats Limit reached${resetInfo}${!terse && '. /auth to continue.'}`}
{terse ? 'Limit reached' : `Limit reached${resetMsg}`}
</Text>
);
}
@@ -57,8 +55,12 @@ export const QuotaDisplay: React.FC<QuotaDisplayProps> = ({
return (
<Text color={color}>
{terse
? `${percentage.toFixed(0)}%`
: `/stats ${percentage.toFixed(0)}% usage remaining${resetInfo}`}
? `${usedPercentage.toFixed(0)}%`
: `${usedPercentage.toFixed(0)}% used${
resetTime
? ` (Limit resets in ${formatResetTime(resetTime, true)})`
: ''
}`}
</Text>
);
};
@@ -9,9 +9,9 @@ import { Box, Text } from 'ink';
import { theme } from '../semantic-colors.js';
import { formatResetTime } from '../utils/formatters.js';
import {
getStatusColor,
QUOTA_THRESHOLD_HIGH,
QUOTA_THRESHOLD_MEDIUM,
getUsedStatusColor,
QUOTA_USED_WARNING_THRESHOLD,
QUOTA_USED_CRITICAL_THRESHOLD,
} from '../utils/displayUtils.js';
interface QuotaStatsInfoProps {
@@ -31,19 +31,24 @@ export const QuotaStatsInfo: React.FC<QuotaStatsInfoProps> = ({
return null;
}
const percentage = (remaining / limit) * 100;
const color = getStatusColor(percentage, {
green: QUOTA_THRESHOLD_HIGH,
yellow: QUOTA_THRESHOLD_MEDIUM,
const usedPercentage = 100 - (remaining / limit) * 100;
const color = getUsedStatusColor(usedPercentage, {
warning: QUOTA_USED_WARNING_THRESHOLD,
critical: QUOTA_USED_CRITICAL_THRESHOLD,
});
return (
<Box flexDirection="column" marginTop={0} marginBottom={0}>
<Text color={color}>
{remaining === 0
? `Limit reached`
: `${percentage.toFixed(0)}% usage remaining`}
{resetTime && `, ${formatResetTime(resetTime)}`}
? `Limit reached${
resetTime ? `, resets in ${formatResetTime(resetTime, true)}` : ''
}`
: `${usedPercentage.toFixed(0)}% used${
resetTime
? ` (Limit resets in ${formatResetTime(resetTime, true)})`
: ''
}`}
</Text>
{showDetails && (
<>
@@ -465,9 +465,9 @@ describe('<StatsDisplay />', () => {
await waitUntilReady();
const output = lastFrame();
expect(output).toContain('Usage remaining');
expect(output).toContain('75.0%');
expect(output).toContain('resets in 1h 30m');
expect(output).toContain('Model usage');
expect(output).toContain('25%');
expect(output).toContain('Usage resets');
expect(output).toMatchSnapshot();
vi.useRealTimers();
@@ -521,8 +521,8 @@ describe('<StatsDisplay />', () => {
await waitUntilReady();
const output = lastFrame();
// (10 + 700) / (100 + 1000) = 710 / 1100 = 64.5%
expect(output).toContain('65% usage remaining');
// (1 - 710/1100) * 100 = 35.5%
expect(output).toContain('35%');
expect(output).toContain('Usage limit: 1,100');
expect(output).toMatchSnapshot();
@@ -571,8 +571,8 @@ describe('<StatsDisplay />', () => {
expect(output).toContain('gemini-2.5-flash');
expect(output).toContain('-'); // for requests
expect(output).toContain('50.0%');
expect(output).toContain('resets in 2h');
expect(output).toContain('50%');
expect(output).toContain('Usage resets');
expect(output).toMatchSnapshot();
vi.useRealTimers();
+164 -40
View File
@@ -5,7 +5,7 @@
*/
import type React from 'react';
import { Box, Text } from 'ink';
import { Box, Text, useStdout } from 'ink';
import { ThemedGradient } from './ThemedGradient.js';
import { theme } from '../semantic-colors.js';
import { formatDuration, formatResetTime } from '../utils/formatters.js';
@@ -19,6 +19,9 @@ import {
USER_AGREEMENT_RATE_MEDIUM,
CACHE_EFFICIENCY_HIGH,
CACHE_EFFICIENCY_MEDIUM,
getUsedStatusColor,
QUOTA_USED_WARNING_THRESHOLD,
QUOTA_USED_CRITICAL_THRESHOLD,
} from '../utils/displayUtils.js';
import { computeSessionStats } from '../utils/computeStats.js';
import {
@@ -155,6 +158,8 @@ const ModelUsageTable: React.FC<{
useGemini3_1,
useCustomToolModel,
}) => {
const { stdout } = useStdout();
const terminalWidth = stdout?.columns ?? 84;
const rows = buildModelRows(models, quotas, useGemini3_1, useCustomToolModel);
if (rows.length === 0) {
@@ -163,12 +168,47 @@ const ModelUsageTable: React.FC<{
const showQuotaColumn = !!quotas && rows.some((row) => !!row.bucket);
const nameWidth = 25;
const requestsWidth = 7;
const nameWidth = 23;
const requestsWidth = 5;
const uncachedWidth = 15;
const cachedWidth = 14;
const outputTokensWidth = 15;
const usageLimitWidth = showQuotaColumn ? 28 : 0;
const percentageWidth = showQuotaColumn ? 6 : 0;
const resetWidth = 22;
// Total width of other columns (including parent box paddingX={2})
const fixedWidth = nameWidth + requestsWidth + percentageWidth + resetWidth;
const outerPadding = 4;
const availableForUsage = terminalWidth - outerPadding - fixedWidth;
const usageLimitWidth = showQuotaColumn
? Math.max(10, Math.min(24, availableForUsage))
: 0;
const progressBarWidth = Math.max(2, usageLimitWidth - 4);
const renderProgressBar = (
usedFraction: number,
color: string,
totalSteps = 20,
) => {
let filledSteps = Math.round(usedFraction * totalSteps);
// If something is used (fraction > 0) but rounds to 0, show 1 tick.
// If < 100% (fraction < 1) but rounds to 20, show 19 ticks.
if (usedFraction > 0 && usedFraction < 1) {
filledSteps = Math.min(Math.max(filledSteps, 1), totalSteps - 1);
}
const emptySteps = Math.max(0, totalSteps - filledSteps);
return (
<Box flexDirection="row" flexShrink={0}>
<Text wrap="truncate-end">
<Text color={color}>{'▬'.repeat(filledSteps)}</Text>
<Text color={theme.border.default}>{'▬'.repeat(emptySteps)}</Text>
</Text>
</Box>
);
};
const cacheEfficiencyColor = getStatusColor(cacheEfficiency, {
green: CACHE_EFFICIENCY_HIGH,
@@ -179,25 +219,13 @@ const ModelUsageTable: React.FC<{
nameWidth +
requestsWidth +
(showQuotaColumn
? usageLimitWidth
? usageLimitWidth + percentageWidth + resetWidth
: uncachedWidth + cachedWidth + outputTokensWidth);
const isAuto = currentModel && isAutoModel(currentModel);
const modelUsageTitle = isAuto
? `${getDisplayString(currentModel)} Usage`
: `Model Usage`;
return (
<Box flexDirection="column" marginBottom={1}>
{/* Header */}
<Box alignItems="flex-end">
<Box width={nameWidth}>
<Text bold color={theme.text.primary} wrap="truncate-end">
{modelUsageTitle}
</Text>
</Box>
</Box>
{isAuto &&
showQuotaColumn &&
pooledRemaining !== undefined &&
@@ -216,7 +244,7 @@ const ModelUsageTable: React.FC<{
)}
<Box alignItems="flex-end">
<Box width={nameWidth}>
<Box width={nameWidth} flexShrink={0}>
<Text bold color={theme.text.primary}>
Model
</Text>
@@ -267,15 +295,31 @@ const ModelUsageTable: React.FC<{
</>
)}
{showQuotaColumn && (
<Box
width={usageLimitWidth}
flexDirection="column"
alignItems="flex-end"
>
<Text bold color={theme.text.primary}>
Usage remaining
</Text>
</Box>
<>
<Box
width={usageLimitWidth}
flexDirection="column"
alignItems="flex-start"
paddingLeft={4}
flexShrink={0}
>
<Text bold color={theme.text.primary}>
Model usage
</Text>
</Box>
<Box width={percentageWidth} flexShrink={0} />
<Box
width={resetWidth}
flexDirection="column"
alignItems="flex-start"
paddingLeft={2}
flexShrink={0}
>
<Text bold color={theme.text.primary} wrap="truncate-end">
Usage resets
</Text>
</Box>
</>
)}
</Box>
@@ -292,7 +336,7 @@ const ModelUsageTable: React.FC<{
{rows.map((row) => (
<Box key={row.key}>
<Box width={nameWidth}>
<Box width={nameWidth} flexShrink={0}>
<Text
color={row.isActive ? theme.text.primary : theme.text.secondary}
wrap="truncate-end"
@@ -352,20 +396,100 @@ const ModelUsageTable: React.FC<{
</Box>
</>
)}
<Box
width={usageLimitWidth}
flexDirection="column"
alignItems="flex-end"
>
{row.bucket &&
row.bucket.remainingFraction != null &&
row.bucket.resetTime && (
{showQuotaColumn && (
<>
<Box
width={usageLimitWidth}
flexDirection="column"
alignItems="flex-start"
paddingLeft={4}
flexShrink={0}
>
{row.bucket && row.bucket.remainingFraction != null && (
<Box flexDirection="row" flexShrink={0}>
{(() => {
const actualUsedFraction =
1 - row.bucket.remainingFraction;
const effectiveUsedFraction =
actualUsedFraction === 0 && row.isActive
? 0.001
: actualUsedFraction;
const usedPercentage = effectiveUsedFraction * 100;
const statusColor =
getUsedStatusColor(usedPercentage, {
warning: QUOTA_USED_WARNING_THRESHOLD,
critical: QUOTA_USED_CRITICAL_THRESHOLD,
}) ??
(row.isActive ? theme.text.primary : theme.ui.comment);
return renderProgressBar(
effectiveUsedFraction,
statusColor,
progressBarWidth,
);
})()}
</Box>
)}
</Box>
<Box
width={percentageWidth}
flexDirection="column"
alignItems="flex-end"
flexShrink={0}
>
{row.bucket && row.bucket.remainingFraction != null && (
<Box>
{(() => {
const actualUsedFraction =
1 - row.bucket.remainingFraction;
const effectiveUsedFraction =
actualUsedFraction === 0 && row.isActive
? 0.001
: actualUsedFraction;
const usedPercentage = effectiveUsedFraction * 100;
const statusColor =
getUsedStatusColor(usedPercentage, {
warning: QUOTA_USED_WARNING_THRESHOLD,
critical: QUOTA_USED_CRITICAL_THRESHOLD,
}) ??
(row.isActive ? theme.text.primary : theme.ui.comment);
const percentageText =
usedPercentage > 0 && usedPercentage < 1
? `${usedPercentage.toFixed(1)}%`
: `${usedPercentage.toFixed(0)}%`;
return row.bucket.remainingFraction === 0 ? (
<Text color={theme.status.error} wrap="truncate-end">
Limit
</Text>
) : (
<Text color={statusColor} wrap="truncate-end">
{percentageText}
</Text>
);
})()}
</Box>
)}
</Box>
<Box
width={resetWidth}
flexDirection="column"
alignItems="flex-start"
paddingLeft={2}
flexShrink={0}
>
<Text color={theme.text.secondary} wrap="truncate-end">
{(row.bucket.remainingFraction * 100).toFixed(1)}%{' '}
{formatResetTime(row.bucket.resetTime)}
{row.bucket?.resetTime &&
formatResetTime(row.bucket.resetTime, 'column')
? formatResetTime(row.bucket.resetTime, 'column')
: ''}
</Text>
)}
</Box>
</Box>
</>
)}
</Box>
))}
@@ -89,11 +89,12 @@ const renderStatusDisplay = async (
};
describe('StatusDisplay', () => {
const originalEnv = process.env;
beforeEach(() => {
vi.stubEnv('GEMINI_SYSTEM_MD', '');
});
afterEach(() => {
process.env = { ...originalEnv };
delete process.env['GEMINI_SYSTEM_MD'];
vi.unstubAllEnvs();
vi.restoreAllMocks();
});
@@ -112,7 +113,7 @@ describe('StatusDisplay', () => {
});
it('renders system md indicator if env var is set', async () => {
process.env['GEMINI_SYSTEM_MD'] = 'true';
vi.stubEnv('GEMINI_SYSTEM_MD', 'true');
const { lastFrame, unmount } = await renderStatusDisplay();
expect(lastFrame()).toMatchSnapshot();
unmount();
@@ -6,7 +6,7 @@ exports[`<Footer /> > displays "Limit reached" message when remaining is 0 1`] =
`;
exports[`<Footer /> > displays the usage indicator when usage is low 1`] = `
" ...directories/to/make/it/long no sandbox (see /docs) /model gemini-pro 15%
" ...directories/to/make/it/long no sandbox (see /docs) /model gemini-pro 85%
"
`;
@@ -1,12 +1,12 @@
// Vitest Snapshot v1, https://vitest.dev/guide/snapshot.html
exports[`QuotaDisplay > should NOT render reset time when terse is true 1`] = `
"15%
"85%
"
`;
exports[`QuotaDisplay > should render red when usage < 5% 1`] = `
"/stats 4% usage remaining
exports[`QuotaDisplay > should render critical when used >= 95% 1`] = `
"96% used
"
`;
@@ -15,12 +15,12 @@ exports[`QuotaDisplay > should render terse limit reached message 1`] = `
"
`;
exports[`QuotaDisplay > should render with reset time when provided 1`] = `
"/stats 15% usage remaining, resets in 1h
exports[`QuotaDisplay > should render warning when used >= 80% 1`] = `
"85% used
"
`;
exports[`QuotaDisplay > should render yellow when usage < 20% 1`] = `
"/stats 15% usage remaining
exports[`QuotaDisplay > should render with reset time when provided 1`] = `
"85% used (Limit resets in 1h)
"
`;
@@ -25,15 +25,15 @@ exports[`SettingsDialog > Initial Rendering > should render settings list with v
│ Plan Directory undefined │
│ The directory where planning artifacts are stored. If not specified, defaults t… │
│ │
│ Max Chat Model Attempts 10 │
│ Maximum number of attempts for requests to the main chat model. Cannot exceed 10. │
│ │
│ Debug Keystroke Logging false │
│ Enable debug logging of keystrokes to the console. │
│ │
│ Enable Session Cleanup false │
│ Enable automatic session cleanup │
│ │
│ Keep chat history undefined │
│ Automatically delete chats older than this time period (e.g., "30d", "7d", "24h… │
│ │
│ ▼ │
│ │
│ Apply To │
@@ -72,15 +72,15 @@ exports[`SettingsDialog > Snapshot Tests > should render 'accessibility settings
│ Plan Directory undefined │
│ The directory where planning artifacts are stored. If not specified, defaults t… │
│ │
│ Max Chat Model Attempts 10 │
│ Maximum number of attempts for requests to the main chat model. Cannot exceed 10. │
│ │
│ Debug Keystroke Logging false │
│ Enable debug logging of keystrokes to the console. │
│ │
│ Enable Session Cleanup false │
│ Enable automatic session cleanup │
│ │
│ Keep chat history undefined │
│ Automatically delete chats older than this time period (e.g., "30d", "7d", "24h… │
│ │
│ ▼ │
│ │
│ Apply To │
@@ -119,15 +119,15 @@ exports[`SettingsDialog > Snapshot Tests > should render 'all boolean settings d
│ Plan Directory undefined │
│ The directory where planning artifacts are stored. If not specified, defaults t… │
│ │
│ Max Chat Model Attempts 10 │
│ Maximum number of attempts for requests to the main chat model. Cannot exceed 10. │
│ │
│ Debug Keystroke Logging false* │
│ Enable debug logging of keystrokes to the console. │
│ │
│ Enable Session Cleanup false │
│ Enable automatic session cleanup │
│ │
│ Keep chat history undefined │
│ Automatically delete chats older than this time period (e.g., "30d", "7d", "24h… │
│ │
│ ▼ │
│ │
│ Apply To │
@@ -166,15 +166,15 @@ exports[`SettingsDialog > Snapshot Tests > should render 'default state' correct
│ Plan Directory undefined │
│ The directory where planning artifacts are stored. If not specified, defaults t… │
│ │
│ Max Chat Model Attempts 10 │
│ Maximum number of attempts for requests to the main chat model. Cannot exceed 10. │
│ │
│ Debug Keystroke Logging false │
│ Enable debug logging of keystrokes to the console. │
│ │
│ Enable Session Cleanup false │
│ Enable automatic session cleanup │
│ │
│ Keep chat history undefined │
│ Automatically delete chats older than this time period (e.g., "30d", "7d", "24h… │
│ │
│ ▼ │
│ │
│ Apply To │
@@ -213,15 +213,15 @@ exports[`SettingsDialog > Snapshot Tests > should render 'file filtering setting
│ Plan Directory undefined │
│ The directory where planning artifacts are stored. If not specified, defaults t… │
│ │
│ Max Chat Model Attempts 10 │
│ Maximum number of attempts for requests to the main chat model. Cannot exceed 10. │
│ │
│ Debug Keystroke Logging false │
│ Enable debug logging of keystrokes to the console. │
│ │
│ Enable Session Cleanup false │
│ Enable automatic session cleanup │
│ │
│ Keep chat history undefined │
│ Automatically delete chats older than this time period (e.g., "30d", "7d", "24h… │
│ │
│ ▼ │
│ │
│ Apply To │
@@ -260,15 +260,15 @@ exports[`SettingsDialog > Snapshot Tests > should render 'focused on scope selec
│ Plan Directory undefined │
│ The directory where planning artifacts are stored. If not specified, defaults t… │
│ │
│ Max Chat Model Attempts 10 │
│ Maximum number of attempts for requests to the main chat model. Cannot exceed 10. │
│ │
│ Debug Keystroke Logging false │
│ Enable debug logging of keystrokes to the console. │
│ │
│ Enable Session Cleanup false │
│ Enable automatic session cleanup │
│ │
│ Keep chat history undefined │
│ Automatically delete chats older than this time period (e.g., "30d", "7d", "24h… │
│ │
│ ▼ │
│ │
│ > Apply To │
@@ -307,15 +307,15 @@ exports[`SettingsDialog > Snapshot Tests > should render 'mixed boolean and numb
│ Plan Directory undefined │
│ The directory where planning artifacts are stored. If not specified, defaults t… │
│ │
│ Max Chat Model Attempts 10 │
│ Maximum number of attempts for requests to the main chat model. Cannot exceed 10. │
│ │
│ Debug Keystroke Logging false │
│ Enable debug logging of keystrokes to the console. │
│ │
│ Enable Session Cleanup false │
│ Enable automatic session cleanup │
│ │
│ Keep chat history undefined │
│ Automatically delete chats older than this time period (e.g., "30d", "7d", "24h… │
│ │
│ ▼ │
│ │
│ Apply To │
@@ -354,15 +354,15 @@ exports[`SettingsDialog > Snapshot Tests > should render 'tools and security set
│ Plan Directory undefined │
│ The directory where planning artifacts are stored. If not specified, defaults t… │
│ │
│ Max Chat Model Attempts 10 │
│ Maximum number of attempts for requests to the main chat model. Cannot exceed 10. │
│ │
│ Debug Keystroke Logging false │
│ Enable debug logging of keystrokes to the console. │
│ │
│ Enable Session Cleanup false │
│ Enable automatic session cleanup │
│ │
│ Keep chat history undefined │
│ Automatically delete chats older than this time period (e.g., "30d", "7d", "24h… │
│ │
│ ▼ │
│ │
│ Apply To │
@@ -401,15 +401,15 @@ exports[`SettingsDialog > Snapshot Tests > should render 'various boolean settin
│ Plan Directory undefined │
│ The directory where planning artifacts are stored. If not specified, defaults t… │
│ │
│ Max Chat Model Attempts 10 │
│ Maximum number of attempts for requests to the main chat model. Cannot exceed 10. │
│ │
│ Debug Keystroke Logging true* │
│ Enable debug logging of keystrokes to the console. │
│ │
│ Enable Session Cleanup false │
│ Enable automatic session cleanup │
│ │
│ Keep chat history undefined │
│ Automatically delete chats older than this time period (e.g., "30d", "7d", "24h… │
│ │
│ ▼ │
│ │
│ Apply To │
@@ -117,10 +117,9 @@ exports[`<StatsDisplay /> > Conditional Rendering Tests > hides Efficiency secti
│ » API Time: 100ms (100.0%) │
│ » Tool Time: 0s (0.0%) │
│ │
│ Model Usage
Model Reqs Input Tokens Cache Reads Output Tokens
────────────────────────────────────────────────────────────────────────────
│ gemini-2.5-pro 1 100 0 100 │
│ Model Reqs Input Tokens Cache Reads Output Tokens
────────────────────────────────────────────────────────────────────────
gemini-2.5-pro 1 100 0 100
│ │
╰──────────────────────────────────────────────────────────────────────────────────────────────────╯
"
@@ -162,16 +161,15 @@ exports[`<StatsDisplay /> > Quota Display > renders pooled quota information for
│ » API Time: 0s (0.0%) │
│ » Tool Time: 0s (0.0%) │
│ │
auto Usage
│ 65% usage remaining │
35% used
│ Usage limit: 1,100 │
│ Usage limits span all sessions and reset daily. │
│ For a full token breakdown, run \`/stats model\`. │
│ │
│ Model Reqs Usage remaining
│ ────────────────────────────────────────────────────────────
│ gemini-2.5-pro -
│ gemini-2.5-flash -
│ Model Reqs Model usage Usage resets
│ ────────────────────────────────────────────────────────────────────────────────
│ gemini-2.5-pro - ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ 90%
│ gemini-2.5-flash - ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ 30%
│ │
╰──────────────────────────────────────────────────────────────────────────────────────────────────╯
"
@@ -193,10 +191,9 @@ exports[`<StatsDisplay /> > Quota Display > renders quota information for unused
│ » API Time: 0s (0.0%) │
│ » Tool Time: 0s (0.0%) │
│ │
│ Model Usage
Model Reqs Usage remaining
────────────────────────────────────────────────────────────
│ gemini-2.5-flash - 50.0% resets in 2h │
│ Model Reqs Model usage Usage resets
────────────────────────────────────────────────────────────────────────────────
gemini-2.5-flash - ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ 50% 6:00 AM (2h)
│ │
╰──────────────────────────────────────────────────────────────────────────────────────────────────╯
"
@@ -218,10 +215,9 @@ exports[`<StatsDisplay /> > Quota Display > renders quota information when quota
│ » API Time: 100ms (100.0%) │
│ » Tool Time: 0s (0.0%) │
│ │
│ Model Usage
Model Reqs Usage remaining
────────────────────────────────────────────────────────────
│ gemini-2.5-pro 1 75.0% resets in 1h 30m │
│ Model Reqs Model usage Usage resets
────────────────────────────────────────────────────────────────────────────────
gemini-2.5-pro 1 ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ 25% 5:30 AM (1h 30m)
│ │
╰──────────────────────────────────────────────────────────────────────────────────────────────────╯
"
@@ -283,11 +279,10 @@ exports[`<StatsDisplay /> > renders a table with two models correctly 1`] = `
│ » API Time: 19.5s (100.0%) │
│ » Tool Time: 0s (0.0%) │
│ │
│ Model Usage
Model Reqs Input Tokens Cache Reads Output Tokens
────────────────────────────────────────────────────────────────────────────
│ gemini-2.5-pro 3 500 500 2,000 │
│ gemini-2.5-flash 5 15,000 10,000 15,000 │
│ Model Reqs Input Tokens Cache Reads Output Tokens
────────────────────────────────────────────────────────────────────────
gemini-2.5-pro 3 500 500 2,000
│ gemini-2.5-flash 5 15,000 10,000 15,000
│ │
│ Savings Highlight: 10,500 (40.4%) of input tokens were served from the cache, reducing costs. │
│ │
@@ -312,10 +307,9 @@ exports[`<StatsDisplay /> > renders all sections when all data is present 1`] =
│ » API Time: 100ms (44.8%) │
│ » Tool Time: 123ms (55.2%) │
│ │
│ Model Usage
Model Reqs Input Tokens Cache Reads Output Tokens
────────────────────────────────────────────────────────────────────────────
│ gemini-2.5-pro 1 50 50 100 │
│ Model Reqs Input Tokens Cache Reads Output Tokens
────────────────────────────────────────────────────────────────────────
gemini-2.5-pro 1 50 50 100
│ │
│ Savings Highlight: 50 (50.0%) of input tokens were served from the cache, reducing costs. │
│ │
@@ -58,7 +58,10 @@ export const ShellToolMessage: React.FC<ShellToolMessageProps> = ({
borderColor,
borderDimColor,
isExpandable,
originalRequestName,
}) => {
const {
activePtyId: activeShellPtyId,
@@ -129,6 +132,7 @@ export const ShellToolMessage: React.FC<ShellToolMessageProps> = ({
status={status}
description={description}
emphasis={emphasis}
originalRequestName={originalRequestName}
/>
<FocusHint
@@ -520,4 +520,77 @@ describe('ToolConfirmationMessage', () => {
expect(output).toMatchSnapshot();
unmount();
});
it('should show MCP tool details expand hint for MCP confirmations', async () => {
const confirmationDetails: ToolCallConfirmationDetails = {
type: 'mcp',
title: 'Confirm MCP Tool',
serverName: 'test-server',
toolName: 'test-tool',
toolDisplayName: 'Test Tool',
toolArgs: {
url: 'https://www.google.co.jp',
},
toolDescription: 'Navigates browser to a URL.',
toolParameterSchema: {
type: 'object',
properties: {
url: {
type: 'string',
description: 'Destination URL',
},
},
required: ['url'],
},
onConfirm: vi.fn(),
};
const { lastFrame, waitUntilReady, unmount } = renderWithProviders(
<ToolConfirmationMessage
callId="test-call-id"
confirmationDetails={confirmationDetails}
config={mockConfig}
availableTerminalHeight={30}
terminalWidth={80}
/>,
);
await waitUntilReady();
const output = lastFrame();
expect(output).toContain('MCP Tool Details:');
expect(output).toContain('(press Ctrl+O to expand MCP tool details)');
expect(output).not.toContain('https://www.google.co.jp');
expect(output).not.toContain('Navigates browser to a URL.');
unmount();
});
it('should omit empty MCP invocation arguments from details', async () => {
const confirmationDetails: ToolCallConfirmationDetails = {
type: 'mcp',
title: 'Confirm MCP Tool',
serverName: 'test-server',
toolName: 'test-tool',
toolDisplayName: 'Test Tool',
toolArgs: {},
toolDescription: 'No arguments required.',
onConfirm: vi.fn(),
};
const { lastFrame, waitUntilReady, unmount } = renderWithProviders(
<ToolConfirmationMessage
callId="test-call-id"
confirmationDetails={confirmationDetails}
config={mockConfig}
availableTerminalHeight={30}
terminalWidth={80}
/>,
);
await waitUntilReady();
const output = lastFrame();
expect(output).toContain('MCP Tool Details:');
expect(output).toContain('(press Ctrl+O to expand MCP tool details)');
expect(output).not.toContain('Invocation Arguments:');
unmount();
});
});
@@ -5,7 +5,7 @@
*/
import type React from 'react';
import { useMemo, useCallback } from 'react';
import { useMemo, useCallback, useState } from 'react';
import { Box, Text } from 'ink';
import { DiffRenderer } from './DiffRenderer.js';
import { RenderInline } from '../../utils/InlineMarkdownRenderer.js';
@@ -29,6 +29,7 @@ import { useKeypress } from '../../hooks/useKeypress.js';
import { theme } from '../../semantic-colors.js';
import { useSettings } from '../../contexts/SettingsContext.js';
import { keyMatchers, Command } from '../../keyMatchers.js';
import { formatCommand } from '../../utils/keybindingUtils.js';
import {
REDIRECTION_WARNING_NOTE_LABEL,
REDIRECTION_WARNING_NOTE_TEXT,
@@ -64,6 +65,17 @@ export const ToolConfirmationMessage: React.FC<
terminalWidth,
}) => {
const { confirm, isDiffingEnabled } = useToolActions();
const [mcpDetailsExpansionState, setMcpDetailsExpansionState] = useState<{
callId: string;
expanded: boolean;
}>({
callId,
expanded: false,
});
const isMcpToolDetailsExpanded =
mcpDetailsExpansionState.callId === callId
? mcpDetailsExpansionState.expanded
: false;
const settings = useSettings();
const allowPermanentApproval =
@@ -86,9 +98,81 @@ export const ToolConfirmationMessage: React.FC<
[confirm, callId],
);
const mcpToolDetailsText = useMemo(() => {
if (confirmationDetails.type !== 'mcp') {
return null;
}
const detailsLines: string[] = [];
const hasNonEmptyToolArgs =
confirmationDetails.toolArgs !== undefined &&
!(
typeof confirmationDetails.toolArgs === 'object' &&
confirmationDetails.toolArgs !== null &&
Object.keys(confirmationDetails.toolArgs).length === 0
);
if (hasNonEmptyToolArgs) {
let argsText: string;
try {
argsText = stripUnsafeCharacters(
JSON.stringify(confirmationDetails.toolArgs, null, 2),
);
} catch {
argsText = '[unserializable arguments]';
}
detailsLines.push('Invocation Arguments:');
detailsLines.push(argsText);
}
const description = confirmationDetails.toolDescription?.trim();
if (description) {
if (detailsLines.length > 0) {
detailsLines.push('');
}
detailsLines.push('Description:');
detailsLines.push(stripUnsafeCharacters(description));
}
if (confirmationDetails.toolParameterSchema !== undefined) {
let schemaText: string;
try {
schemaText = stripUnsafeCharacters(
JSON.stringify(confirmationDetails.toolParameterSchema, null, 2),
);
} catch {
schemaText = '[unserializable schema]';
}
if (detailsLines.length > 0) {
detailsLines.push('');
}
detailsLines.push('Input Schema:');
detailsLines.push(schemaText);
}
if (detailsLines.length === 0) {
return null;
}
return detailsLines.join('\n');
}, [confirmationDetails]);
const hasMcpToolDetails = !!mcpToolDetailsText;
const expandDetailsHintKey = formatCommand(Command.SHOW_MORE_LINES);
useKeypress(
(key) => {
if (!isFocused) return false;
if (
confirmationDetails.type === 'mcp' &&
hasMcpToolDetails &&
keyMatchers[Command.SHOW_MORE_LINES](key)
) {
setMcpDetailsExpansionState({
callId,
expanded: !isMcpToolDetailsExpanded,
});
return true;
}
if (keyMatchers[Command.ESCAPE](key)) {
handleConfirm(ToolConfirmationOutcome.Cancel);
return true;
@@ -100,7 +184,7 @@ export const ToolConfirmationMessage: React.FC<
}
return false;
},
{ isActive: isFocused },
{ isActive: isFocused, priority: true },
);
const handleSelect = useCallback(
@@ -504,12 +588,31 @@ export const ToolConfirmationMessage: React.FC<
bodyContent = (
<Box flexDirection="column">
<Text color={theme.text.link}>
MCP Server: {sanitizeForDisplay(mcpProps.serverName)}
</Text>
<Text color={theme.text.link}>
Tool: {sanitizeForDisplay(mcpProps.toolName)}
</Text>
<>
<Text color={theme.text.link}>
MCP Server: {sanitizeForDisplay(mcpProps.serverName)}
</Text>
<Text color={theme.text.link}>
Tool: {sanitizeForDisplay(mcpProps.toolName)}
</Text>
</>
{hasMcpToolDetails && (
<Box flexDirection="column" marginTop={1}>
<Text color={theme.text.primary}>MCP Tool Details:</Text>
{isMcpToolDetailsExpanded ? (
<>
<Text color={theme.text.secondary}>
(press {expandDetailsHintKey} to collapse MCP tool details)
</Text>
<Text color={theme.text.link}>{mcpToolDetailsText}</Text>
</>
) : (
<Text color={theme.text.secondary}>
(press {expandDetailsHintKey} to expand MCP tool details)
</Text>
)}
</Box>
)}
</Box>
);
}
@@ -522,8 +625,17 @@ export const ToolConfirmationMessage: React.FC<
terminalWidth,
handleConfirm,
deceptiveUrlWarningText,
isMcpToolDetailsExpanded,
hasMcpToolDetails,
mcpToolDetailsText,
expandDetailsHintKey,
]);
const bodyOverflowDirection: 'top' | 'bottom' =
confirmationDetails.type === 'mcp' && isMcpToolDetailsExpanded
? 'bottom'
: 'top';
if (confirmationDetails.type === 'edit') {
if (confirmationDetails.isModifying) {
return (
@@ -559,7 +671,7 @@ export const ToolConfirmationMessage: React.FC<
<MaxSizedBox
maxHeight={availableBodyContentHeight()}
maxWidth={terminalWidth}
overflowDirection="top"
overflowDirection={bodyOverflowDirection}
>
{bodyContent}
</MaxSizedBox>
@@ -57,6 +57,7 @@ export const ToolMessage: React.FC<ToolMessageProps> = ({
config,
progressMessage,
progressPercent,
originalRequestName,
}) => {
const isThisShellFocused = checkIsShellFocused(
name,
@@ -93,6 +94,7 @@ export const ToolMessage: React.FC<ToolMessageProps> = ({
emphasis={emphasis}
progressMessage={progressMessage}
progressPercent={progressPercent}
originalRequestName={originalRequestName}
/>
<FocusHint
shouldShowFocusHint={shouldShowFocusHint}
@@ -189,6 +189,7 @@ type ToolInfoProps = {
emphasis: TextEmphasis;
progressMessage?: string;
progressPercent?: number;
originalRequestName?: string;
};
export const ToolInfo: React.FC<ToolInfoProps> = ({
@@ -198,6 +199,7 @@ export const ToolInfo: React.FC<ToolInfoProps> = ({
emphasis,
progressMessage,
progressPercent,
originalRequestName,
}) => {
const status = mapCoreStatusToDisplayStatus(coreStatus);
const nameColor = React.useMemo<string>(() => {
@@ -242,6 +244,12 @@ export const ToolInfo: React.FC<ToolInfoProps> = ({
<Text color={nameColor} bold>
{name}
</Text>
{originalRequestName && originalRequestName !== name && (
<Text color={theme.text.secondary} italic>
{' '}
(redirection from {originalRequestName})
</Text>
)}
{!isCompletedAskUser && (
<>
{' '}
@@ -275,5 +275,20 @@ describe('toolMapping', () => {
expect(result.tools[0].resultDisplay).toBeUndefined();
expect(result.tools[0].status).toBe(CoreToolCallStatus.Scheduled);
});
it('propagates originalRequestName correctly', () => {
const toolCall: ScheduledToolCall = {
status: CoreToolCallStatus.Scheduled,
request: {
...mockRequest,
originalRequestName: 'original_tool',
},
tool: mockTool,
invocation: mockInvocation,
};
const result = mapToDisplay(toolCall);
expect(result.tools[0].originalRequestName).toBe('original_tool');
});
});
});
+1
View File
@@ -107,6 +107,7 @@ export function mapToDisplay(
progressMessage,
progressPercent,
approvalMode: call.approvalMode,
originalRequestName: call.request.originalRequestName,
};
});
@@ -13,6 +13,7 @@ import {
Scheduler,
type Config,
type MessageBus,
type ExecutingToolCall,
type CompletedToolCall,
type ToolCallsUpdateMessage,
type AnyDeclarativeTool,
@@ -110,7 +111,7 @@ describe('useToolScheduler', () => {
tool: createMockTool(),
invocation: createMockInvocation(),
liveOutput: 'Loading...',
};
} as ExecutingToolCall;
act(() => {
void mockMessageBus.publish({
@@ -405,4 +406,62 @@ describe('useToolScheduler', () => {
toolCalls.find((t) => t.request.callId === 'call-sub')?.schedulerId,
).toBe('subagent-1');
});
it('adapts success/error status to executing when a tail call is present', () => {
vi.useFakeTimers();
const { result } = renderHook(() =>
useToolScheduler(
vi.fn().mockResolvedValue(undefined),
mockConfig,
() => undefined,
),
);
const startTime = Date.now();
vi.advanceTimersByTime(1000);
const mockToolCall = {
status: CoreToolCallStatus.Success as const,
request: {
callId: 'call-1',
name: 'test_tool',
args: {},
isClientInitiated: false,
prompt_id: 'p1',
},
tool: createMockTool(),
invocation: createMockInvocation(),
response: {
callId: 'call-1',
resultDisplay: 'OK',
responseParts: [],
error: undefined,
errorType: undefined,
},
tailToolCallRequest: {
name: 'tail_tool',
args: {},
isClientInitiated: false,
prompt_id: '123',
},
};
act(() => {
void mockMessageBus.publish({
type: MessageBusType.TOOL_CALLS_UPDATE,
toolCalls: [mockToolCall],
schedulerId: ROOT_SCHEDULER_ID,
} as ToolCallsUpdateMessage);
});
const [toolCalls, , , , , lastOutputTime] = result.current;
// Check if status has been adapted to 'executing'
expect(toolCalls[0].status).toBe(CoreToolCallStatus.Executing);
// Check if lastOutputTime was updated due to the transitional state
expect(lastOutputTime).toBeGreaterThan(startTime);
vi.useRealTimers();
});
});
+26 -2
View File
@@ -14,6 +14,7 @@ import {
Scheduler,
type EditorType,
type ToolCallsUpdateMessage,
CoreToolCallStatus,
} from '@google/gemini-cli-core';
import { useCallback, useState, useMemo, useEffect, useRef } from 'react';
@@ -115,7 +116,16 @@ export function useToolScheduler(
useEffect(() => {
const handler = (event: ToolCallsUpdateMessage) => {
// Update output timer for UI spinners (Side Effect)
if (event.toolCalls.some((tc) => tc.status === 'executing')) {
const hasExecuting = event.toolCalls.some(
(tc) =>
tc.status === CoreToolCallStatus.Executing ||
((tc.status === CoreToolCallStatus.Success ||
tc.status === CoreToolCallStatus.Error) &&
'tailToolCallRequest' in tc &&
tc.tailToolCallRequest != null),
);
if (hasExecuting) {
setLastToolOutputTime(Date.now());
}
@@ -238,9 +248,23 @@ function adaptToolCalls(
const prev = prevMap.get(coreCall.request.callId);
const responseSubmittedToGemini = prev?.responseSubmittedToGemini ?? false;
let status = coreCall.status;
// If a tool call has completed but scheduled a tail call, it is in a transitional
// state. Force the UI to render it as "executing".
if (
(status === CoreToolCallStatus.Success ||
status === CoreToolCallStatus.Error) &&
'tailToolCallRequest' in coreCall &&
coreCall.tailToolCallRequest != null
) {
status = CoreToolCallStatus.Executing;
}
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
return {
...coreCall,
status,
responseSubmittedToGemini,
};
} as TrackedToolCall;
});
}
-1
View File
@@ -352,7 +352,6 @@ describe('keyMatchers', () => {
createKey('l', { ctrl: true }),
],
},
// Shell commands
{
command: Command.REVERSE_SEARCH,
+1
View File
@@ -110,6 +110,7 @@ export interface IndividualToolCallDisplay {
approvalMode?: ApprovalMode;
progressMessage?: string;
progressPercent?: number;
originalRequestName?: string;
}
export interface CompressionProps {
+19
View File
@@ -19,6 +19,9 @@ export const CACHE_EFFICIENCY_MEDIUM = 15;
export const QUOTA_THRESHOLD_HIGH = 20;
export const QUOTA_THRESHOLD_MEDIUM = 5;
export const QUOTA_USED_WARNING_THRESHOLD = 80;
export const QUOTA_USED_CRITICAL_THRESHOLD = 95;
// --- Color Logic ---
export const getStatusColor = (
value: number,
@@ -36,3 +39,19 @@ export const getStatusColor = (
}
return options.defaultColor ?? theme.status.error;
};
/**
* Gets the status color based on "used" percentage (where higher is worse).
*/
export const getUsedStatusColor = (
usedPercentage: number,
thresholds: { warning: number; critical: number },
) => {
if (usedPercentage >= thresholds.critical) {
return theme.status.error;
}
if (usedPercentage >= thresholds.warning) {
return theme.status.warning;
}
return undefined;
};
@@ -10,9 +10,46 @@ import {
formatBytes,
formatTimeAgo,
stripReferenceContent,
formatResetTime,
} from './formatters.js';
describe('formatters', () => {
describe('formatResetTime', () => {
const NOW = new Date('2025-01-01T12:00:00Z');
beforeEach(() => {
vi.useFakeTimers();
vi.setSystemTime(NOW);
});
afterEach(() => {
vi.useRealTimers();
});
it('should format full time correctly', () => {
const resetTime = new Date(NOW.getTime() + 90 * 60 * 1000).toISOString(); // 1h 30m
const result = formatResetTime(resetTime);
expect(result).toMatch(/1 hour 30 minutes at \d{1,2}:\d{2} [AP]M/);
});
it('should format terse time correctly', () => {
const resetTime = new Date(NOW.getTime() + 90 * 60 * 1000).toISOString(); // 1h 30m
expect(formatResetTime(resetTime, 'terse')).toBe('1h 30m');
expect(formatResetTime(resetTime, true)).toBe('1h 30m');
});
it('should format column time correctly', () => {
const resetTime = new Date(NOW.getTime() + 90 * 60 * 1000).toISOString(); // 1h 30m
const result = formatResetTime(resetTime, 'column');
expect(result).toMatch(/\d{1,2}:\d{2} [AP]M \(1h 30m\)/);
});
it('should handle zero or negative diff by returning empty string', () => {
const resetTime = new Date(NOW.getTime() - 1000).toISOString();
expect(formatResetTime(resetTime)).toBe('');
});
});
describe('formatBytes', () => {
it('should format bytes into KB', () => {
expect(formatBytes(12345)).toBe('12.1 KB');
+45 -13
View File
@@ -98,26 +98,58 @@ export function stripReferenceContent(text: string): string {
return text.replace(pattern, '').trim();
}
export const formatResetTime = (resetTime: string): string => {
const diff = new Date(resetTime).getTime() - Date.now();
export const formatResetTime = (
resetTime: string | undefined,
format: 'terse' | 'column' | 'full' | boolean = false,
): string => {
if (!resetTime) return '';
const resetDate = new Date(resetTime);
if (isNaN(resetDate.getTime())) return '';
const diff = resetDate.getTime() - Date.now();
if (diff <= 0) return '';
const totalMinutes = Math.ceil(diff / (1000 * 60));
const hours = Math.floor(totalMinutes / 60);
const minutes = totalMinutes % 60;
const fmt = (val: number, unit: 'hour' | 'minute') =>
new Intl.NumberFormat('en', {
style: 'unit',
unit,
unitDisplay: 'narrow',
}).format(val);
const isTerse = format === 'terse' || format === true;
const isColumn = format === 'column';
if (hours > 0 && minutes > 0) {
return `resets in ${fmt(hours, 'hour')} ${fmt(minutes, 'minute')}`;
} else if (hours > 0) {
return `resets in ${fmt(hours, 'hour')}`;
if (isTerse || isColumn) {
const hoursStr = hours > 0 ? `${hours}h` : '';
const minutesStr = minutes > 0 ? `${minutes}m` : '';
const duration =
hoursStr && minutesStr
? `${hoursStr} ${minutesStr}`
: hoursStr || minutesStr;
if (isColumn) {
const timeStr = new Intl.DateTimeFormat('en-US', {
hour: 'numeric',
minute: 'numeric',
}).format(resetDate);
return duration ? `${timeStr} (${duration})` : timeStr;
}
return duration;
}
return `resets in ${fmt(minutes, 'minute')}`;
let duration = '';
if (hours > 0) {
duration = `${hours} hour${hours > 1 ? 's' : ''}`;
if (minutes > 0) {
duration += ` ${minutes} minute${minutes > 1 ? 's' : ''}`;
}
} else {
duration = `${minutes} minute${minutes > 1 ? 's' : ''}`;
}
const timeStr = new Intl.DateTimeFormat('en-US', {
hour: 'numeric',
minute: 'numeric',
timeZoneName: 'short',
}).format(resetDate);
return `${duration} at ${timeStr}`;
};
+130 -26
View File
@@ -22,13 +22,82 @@ import WebSocket from 'ws';
const ACTIVITY_ID_HEADER = 'x-activity-request-id';
const MAX_BUFFER_SIZE = 100;
/** Type guard: Array.isArray doesn't narrow readonly arrays in TS 5.8 */
function isHeaderRecord(
h: http.OutgoingHttpHeaders | readonly string[],
): h is http.OutgoingHttpHeaders {
return !Array.isArray(h);
}
function isRequestOptions(value: unknown): value is http.RequestOptions {
return (
typeof value === 'object' &&
value !== null &&
!(value instanceof URL) &&
!Array.isArray(value)
);
}
function isIncomingMessageCallback(
value: unknown,
): value is (res: http.IncomingMessage) => void {
return typeof value === 'function';
}
type HttpRequestArgs =
| []
| [
url: string | URL | http.RequestOptions,
options?: http.RequestOptions | ((res: http.IncomingMessage) => void),
callback?: (res: http.IncomingMessage) => void,
];
function callHttpRequest(
originalFn: typeof http.request,
args: HttpRequestArgs,
): http.ClientRequest {
if (args.length === 0) {
return originalFn({});
}
if (args.length === 1) {
const first = args[0];
if (typeof first === 'string' || first instanceof URL) {
return originalFn(first);
}
if (isRequestOptions(first)) {
return originalFn(first);
}
return originalFn({});
}
if (args.length === 2) {
const first = args[0];
const second = args[1];
if (typeof first === 'string' || first instanceof URL) {
if (isIncomingMessageCallback(second)) {
return originalFn(first, second);
}
if (isRequestOptions(second)) {
return originalFn(first, second);
}
}
if (isRequestOptions(first) && isIncomingMessageCallback(second)) {
return originalFn(first, second);
}
}
if (args.length === 3) {
const first = args[0];
const second = args[1];
const third = args[2];
if (
(typeof first === 'string' || first instanceof URL) &&
isRequestOptions(second) &&
isIncomingMessageCallback(third)
) {
return originalFn(first, second, third);
}
}
return originalFn({});
}
export interface NetworkLog {
id: string;
timestamp: number;
@@ -364,7 +433,7 @@ export class ActivityLogger extends EventEmitter {
const wrapRequest = (
originalFn: typeof http.request,
args: unknown[],
args: HttpRequestArgs,
protocol: string,
) => {
const firstArg = args[0];
@@ -373,8 +442,10 @@ export class ActivityLogger extends EventEmitter {
options = firstArg;
} else if (firstArg instanceof URL) {
options = firstArg;
} else if (firstArg && typeof firstArg === 'object') {
options = isRequestOptions(firstArg) ? firstArg : {};
} else {
options = (firstArg ?? {}) as http.RequestOptions;
options = {};
}
let url = '';
@@ -393,9 +464,9 @@ export class ActivityLogger extends EventEmitter {
`${protocol}//${options.hostname || options.host || 'localhost'}${options.path || '/'}`;
}
if (url.includes('127.0.0.1') || url.includes('localhost'))
// eslint-disable-next-line @typescript-eslint/no-explicit-any, @typescript-eslint/no-unsafe-type-assertion
return originalFn.apply(http, args as any);
if (url.includes('127.0.0.1') || url.includes('localhost')) {
return callHttpRequest(originalFn, args);
}
const rawHeaders =
typeof options === 'object' &&
@@ -410,24 +481,23 @@ export class ActivityLogger extends EventEmitter {
if (headers[ACTIVITY_ID_HEADER]) {
delete headers[ACTIVITY_ID_HEADER];
// eslint-disable-next-line @typescript-eslint/no-explicit-any, @typescript-eslint/no-unsafe-type-assertion
return originalFn.apply(http, args as any);
return callHttpRequest(originalFn, args);
}
const id = Math.random().toString(36).substring(7);
this.requestStartTimes.set(id, Date.now());
// eslint-disable-next-line @typescript-eslint/no-explicit-any, @typescript-eslint/no-unsafe-type-assertion
const req = originalFn.apply(http, args as any);
const req = callHttpRequest(originalFn, args);
const requestChunks: Buffer[] = [];
const oldWrite = req.write;
const oldEnd = req.end;
req.write = function (chunk: unknown, ...etc: unknown[]) {
req.write = function (chunk: string | Uint8Array, ...etc: unknown[]) {
if (chunk) {
const encoding =
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
typeof etc[0] === 'string' ? (etc[0] as BufferEncoding) : undefined;
typeof etc[0] === 'string' && Buffer.isEncoding(etc[0])
? etc[0]
: undefined;
requestChunks.push(
Buffer.isBuffer(chunk)
? chunk
@@ -438,19 +508,21 @@ export class ActivityLogger extends EventEmitter {
),
);
}
// eslint-disable-next-line @typescript-eslint/no-explicit-any, @typescript-eslint/no-unsafe-type-assertion
return oldWrite.apply(this, [chunk, ...etc] as any);
// eslint-disable-next-line @typescript-eslint/no-explicit-any, @typescript-eslint/no-unsafe-type-assertion, @typescript-eslint/no-unsafe-return
return (oldWrite as any).apply(this, [chunk, ...etc]);
};
req.end = function (
this: http.ClientRequest,
chunk: unknown,
chunkOrCb?: string | Uint8Array | (() => void),
...etc: unknown[]
) {
if (chunk && typeof chunk !== 'function') {
const chunk = typeof chunkOrCb === 'function' ? undefined : chunkOrCb;
if (chunk) {
const encoding =
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
typeof etc[0] === 'string' ? (etc[0] as BufferEncoding) : undefined;
typeof etc[0] === 'string' && Buffer.isEncoding(etc[0])
? etc[0]
: undefined;
requestChunks.push(
Buffer.isBuffer(chunk)
? chunk
@@ -473,7 +545,7 @@ export class ActivityLogger extends EventEmitter {
pending: true,
});
// eslint-disable-next-line @typescript-eslint/no-explicit-any, @typescript-eslint/no-unsafe-type-assertion, @typescript-eslint/no-unsafe-return
return (oldEnd as any).apply(this, [chunk, ...etc]);
return (oldEnd as any).apply(this, [chunkOrCb, ...etc]);
};
req.on('response', (res: http.IncomingMessage) => {
@@ -545,12 +617,44 @@ export class ActivityLogger extends EventEmitter {
return req;
};
// eslint-disable-next-line @typescript-eslint/no-explicit-any, @typescript-eslint/no-unsafe-type-assertion
(http as any).request = (...args: unknown[]) =>
wrapRequest(originalRequest, args, 'http:');
// eslint-disable-next-line @typescript-eslint/no-explicit-any, @typescript-eslint/no-unsafe-type-assertion
(https as any).request = (...args: unknown[]) =>
wrapRequest(originalHttpsRequest as typeof http.request, args, 'https:');
Object.defineProperty(http, 'request', {
value: (
url: string | URL | http.RequestOptions,
options?: http.RequestOptions | ((res: http.IncomingMessage) => void),
callback?: (res: http.IncomingMessage) => void,
): http.ClientRequest => {
const args: HttpRequestArgs =
callback !== undefined
? [url, options, callback]
: options !== undefined
? [url, options]
: [url];
return wrapRequest(originalRequest, args, 'http:');
},
writable: true,
configurable: true,
});
Object.defineProperty(https, 'request', {
value: (
url: string | URL | http.RequestOptions,
options?: http.RequestOptions | ((res: http.IncomingMessage) => void),
callback?: (res: http.IncomingMessage) => void,
): http.ClientRequest => {
const args: HttpRequestArgs =
callback !== undefined
? [url, options, callback]
: options !== undefined
? [url, options]
: [url];
return wrapRequest(
originalHttpsRequest as typeof http.request,
args,
'https:',
);
},
writable: true,
configurable: true,
});
}
logConsole(payload: ConsoleLogPayload) {
+2
View File
@@ -53,6 +53,8 @@
"ajv-formats": "^3.0.0",
"chardet": "^2.1.0",
"diff": "^8.0.3",
"dotenv": "^17.2.4",
"dotenv-expand": "^12.0.3",
"fast-levenshtein": "^2.0.6",
"fdir": "^6.4.6",
"fzf": "^0.5.2",
@@ -4,13 +4,21 @@
* SPDX-License-Identifier: Apache-2.0
*/
import { describe, it, expect, vi } from 'vitest';
import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest';
import { GeneralistAgent } from './generalist-agent.js';
import { makeFakeConfig } from '../test-utils/config.js';
import type { ToolRegistry } from '../tools/tool-registry.js';
import type { AgentRegistry } from './registry.js';
describe('GeneralistAgent', () => {
beforeEach(() => {
vi.stubEnv('GEMINI_SYSTEM_MD', '');
});
afterEach(() => {
vi.unstubAllEnvs();
});
it('should create a valid generalist agent definition', () => {
const config = makeFakeConfig();
vi.spyOn(config, 'getToolRegistry').mockReturnValue({
@@ -47,7 +47,10 @@ describe('Fallback Integration', () => {
const requestedModel = PREVIEW_GEMINI_MODEL;
// 3. Apply model selection
const result = applyModelSelection(config, { model: requestedModel });
const result = applyModelSelection(config, {
model: requestedModel,
isChatModel: true,
});
// 4. Expect fallback to Flash
expect(result.model).toBe(PREVIEW_GEMINI_FLASH_MODEL);
@@ -222,7 +222,10 @@ describe('policyHelpers', () => {
selectedModel: 'gemini-pro',
});
const result = applyModelSelection(config, { model: 'gemini-pro' });
const result = applyModelSelection(config, {
model: 'gemini-pro',
isChatModel: true,
});
expect(result.model).toBe('gemini-pro');
expect(result.maxAttempts).toBeUndefined();
expect(config.setActiveModel).toHaveBeenCalledWith('gemini-pro');
@@ -243,7 +246,10 @@ describe('policyHelpers', () => {
selectedModel: 'gemini-flash',
});
const result = applyModelSelection(config, { model: 'gemini-pro' });
const result = applyModelSelection(config, {
model: 'gemini-pro',
isChatModel: true,
});
expect(result.model).toBe('gemini-flash');
expect(result.config).toEqual({
@@ -253,14 +259,33 @@ describe('policyHelpers', () => {
expect(mockModelConfigService.getResolvedConfig).toHaveBeenCalledWith({
model: 'gemini-pro',
isChatModel: true,
});
expect(mockModelConfigService.getResolvedConfig).toHaveBeenCalledWith({
model: 'gemini-flash',
isChatModel: true,
});
expect(config.setActiveModel).toHaveBeenCalledWith('gemini-flash');
});
it('consumes sticky attempt if indicated', () => {
it('does not call setActiveModel if isChatModel is false', () => {
const config = createExtendedMockConfig();
mockModelConfigService.getResolvedConfig.mockReturnValue({
model: 'gemini-pro',
generateContentConfig: {},
});
mockAvailabilityService.selectFirstAvailable.mockReturnValue({
selectedModel: 'gemini-pro',
});
applyModelSelection(config, {
model: 'gemini-pro',
isChatModel: false,
});
expect(config.setActiveModel).not.toHaveBeenCalled();
});
it('consumes sticky attempt if indicated and isChatModel is true', () => {
const config = createExtendedMockConfig();
mockModelConfigService.getResolvedConfig.mockReturnValue({
model: 'gemini-pro',
@@ -271,10 +296,36 @@ describe('policyHelpers', () => {
attempts: 1,
});
const result = applyModelSelection(config, { model: 'gemini-pro' });
const result = applyModelSelection(config, {
model: 'gemini-pro',
isChatModel: true,
});
expect(mockAvailabilityService.consumeStickyAttempt).toHaveBeenCalledWith(
'gemini-pro',
);
expect(config.setActiveModel).toHaveBeenCalledWith('gemini-pro');
expect(result.maxAttempts).toBe(1);
});
it('consumes sticky attempt if indicated but does not call setActiveModel if isChatModel is false', () => {
const config = createExtendedMockConfig();
mockModelConfigService.getResolvedConfig.mockReturnValue({
model: 'gemini-pro',
generateContentConfig: {},
});
mockAvailabilityService.selectFirstAvailable.mockReturnValue({
selectedModel: 'gemini-pro',
attempts: 1,
});
const result = applyModelSelection(config, {
model: 'gemini-pro',
isChatModel: false,
});
expect(mockAvailabilityService.consumeStickyAttempt).toHaveBeenCalledWith(
'gemini-pro',
);
expect(config.setActiveModel).not.toHaveBeenCalled();
expect(result.maxAttempts).toBe(1);
});
@@ -291,7 +342,7 @@ describe('policyHelpers', () => {
const result = applyModelSelection(
config,
{ model: 'gemini-pro' },
{ model: 'gemini-pro', isChatModel: true },
{
consumeAttempt: false,
},
@@ -299,6 +350,7 @@ describe('policyHelpers', () => {
expect(
mockAvailabilityService.consumeStickyAttempt,
).not.toHaveBeenCalled();
expect(config.setActiveModel).toHaveBeenCalledWith('gemini-pro');
expect(result.maxAttempts).toBe(1);
});
});
@@ -214,7 +214,9 @@ export function applyModelSelection(
generateContentConfig = fallbackResolved.generateContentConfig;
}
config.setActiveModel(finalModel);
if (modelConfigKey.isChatModel) {
config.setActiveModel(finalModel);
}
if (selection.attempts && options.consumeAttempt !== false) {
config.getModelAvailabilityService().consumeStickyAttempt(finalModel);
+24
View File
@@ -8,6 +8,7 @@ import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest';
import type { Mock } from 'vitest';
import type { ConfigParameters, SandboxConfig } from './config.js';
import { Config, DEFAULT_FILE_FILTERING_OPTIONS } from './config.js';
import { DEFAULT_MAX_ATTEMPTS } from '../utils/retry.js';
import { ExperimentFlags } from '../code_assist/experiments/flagNames.js';
import { debugLogger } from '../utils/debugLogger.js';
import { ApprovalMode } from '../policy/types.js';
@@ -259,6 +260,29 @@ describe('Server Config (config.ts)', () => {
usageStatisticsEnabled: false,
};
describe('maxAttempts', () => {
it('should default to DEFAULT_MAX_ATTEMPTS', () => {
const config = new Config(baseParams);
expect(config.getMaxAttempts()).toBe(DEFAULT_MAX_ATTEMPTS);
});
it('should use provided maxAttempts if <= DEFAULT_MAX_ATTEMPTS', () => {
const config = new Config({
...baseParams,
maxAttempts: 5,
});
expect(config.getMaxAttempts()).toBe(5);
});
it('should cap maxAttempts at DEFAULT_MAX_ATTEMPTS', () => {
const config = new Config({
...baseParams,
maxAttempts: 20,
});
expect(config.getMaxAttempts()).toBe(DEFAULT_MAX_ATTEMPTS);
});
});
beforeEach(() => {
// Reset mocks if necessary
vi.clearAllMocks();
+44 -4
View File
@@ -132,6 +132,11 @@ import { UserHintService } from './userHintService.js';
import { WORKSPACE_POLICY_TIER } from '../policy/config.js';
import { loadPoliciesFromToml } from '../policy/toml-loader.js';
import { CheckerRunner } from '../safety/checker-runner.js';
import { ContextBuilder } from '../safety/context-builder.js';
import { CheckerRegistry } from '../safety/registry.js';
import { ConsecaSafetyChecker } from '../safety/conseca/conseca.js';
export interface AccessibilitySettings {
/** @deprecated Use ui.loadingPhrases instead. */
enableLoadingPhrases?: boolean;
@@ -291,6 +296,7 @@ export interface ExtensionInstallMetadata {
allowPreRelease?: boolean;
}
import { DEFAULT_MAX_ATTEMPTS } from '../utils/retry.js';
import type { FileFilteringOptions } from './constants.js';
import {
DEFAULT_FILE_FILTERING_OPTIONS,
@@ -476,6 +482,7 @@ export interface ConfigParameters {
disableModelRouterForAuth?: AuthType[];
continueOnFailedApiCall?: boolean;
retryFetchErrors?: boolean;
maxAttempts?: number;
enableShellOutputEfficiency?: boolean;
shellToolInactivityTimeout?: number;
fakeResponses?: string;
@@ -511,6 +518,7 @@ export interface ConfigParameters {
adminSkillsEnabled?: boolean;
agents?: AgentSettings;
}>;
enableConseca?: boolean;
}
export class Config {
@@ -538,6 +546,7 @@ export class Config {
private workspaceContext: WorkspaceContext;
private readonly debugMode: boolean;
private readonly question: string | undefined;
readonly enableConseca: boolean;
private readonly coreTools: string[] | undefined;
/** @deprecated Use Policy Engine instead */
@@ -657,6 +666,7 @@ export class Config {
private readonly outputSettings: OutputSettings;
private readonly continueOnFailedApiCall: boolean;
private readonly retryFetchErrors: boolean;
private readonly maxAttempts: number;
private readonly enableShellOutputEfficiency: boolean;
private readonly shellToolInactivityTimeout: number;
readonly fakeResponses?: string;
@@ -865,13 +875,35 @@ export class Config {
this.recordResponses = params.recordResponses;
this.fileExclusions = new FileExclusions(this);
this.eventEmitter = params.eventEmitter;
this.policyEngine = new PolicyEngine({
...params.policyEngineConfig,
approvalMode:
params.approvalMode ?? params.policyEngineConfig?.approvalMode,
this.enableConseca = params.enableConseca ?? false;
// Initialize Safety Infrastructure
const contextBuilder = new ContextBuilder(this);
const checkersPath = this.targetDir;
// The checkersPath is used to resolve external checkers. Since we do not have any external checkers currently, it is set to the targetDir.
const checkerRegistry = new CheckerRegistry(checkersPath);
const checkerRunner = new CheckerRunner(contextBuilder, checkerRegistry, {
checkersPath,
timeout: 30000, // 30 seconds to allow for LLM-based checkers
});
this.policyUpdateConfirmationRequest =
params.policyUpdateConfirmationRequest;
this.policyEngine = new PolicyEngine(
{
...params.policyEngineConfig,
approvalMode:
params.approvalMode ?? params.policyEngineConfig?.approvalMode,
},
checkerRunner,
);
// Register Conseca if enabled
if (this.enableConseca) {
debugLogger.log('[SAFETY] Registering Conseca Safety Checker');
ConsecaSafetyChecker.getInstance().setConfig(this);
}
this.messageBus = new MessageBus(this.policyEngine, this.debugMode);
this.acknowledgedAgentsService = new AcknowledgedAgentsService();
this.skillManager = new SkillManager();
@@ -879,6 +911,10 @@ export class Config {
format: params.output?.format ?? OutputFormat.TEXT,
};
this.retryFetchErrors = params.retryFetchErrors ?? false;
this.maxAttempts = Math.min(
params.maxAttempts ?? DEFAULT_MAX_ATTEMPTS,
DEFAULT_MAX_ATTEMPTS,
);
this.disableYoloMode = params.disableYoloMode ?? false;
this.rawOutput = params.rawOutput ?? false;
this.acceptRawOutputRisk = params.acceptRawOutputRisk ?? false;
@@ -2415,6 +2451,10 @@ export class Config {
return this.retryFetchErrors;
}
getMaxAttempts(): number {
return this.maxAttempts;
}
getEnableShellOutputEfficiency(): boolean {
return this.enableShellOutputEfficiency;
}
+12
View File
@@ -14,6 +14,7 @@ import {
GOOGLE_ACCOUNTS_FILENAME,
isSubpath,
resolveToRealPath,
normalizePath,
} from '../utils/paths.js';
import { ProjectRegistry } from './projectRegistry.js';
import { StorageMigration } from './storageMigration.js';
@@ -142,6 +143,17 @@ export class Storage {
return path.join(this.targetDir, GEMINI_DIR);
}
/**
* Checks if the current workspace storage location is the same as the global/user storage location.
* This handles symlinks and platform-specific path normalization.
*/
isWorkspaceHomeDir(): boolean {
return (
normalizePath(resolveToRealPath(this.targetDir)) ===
normalizePath(resolveToRealPath(homedir()))
);
}
getAgentsDir(): string {
return path.join(this.targetDir, AGENTS_DIR_NAME);
}
@@ -95,6 +95,9 @@ export type SerializableConfirmationDetails =
serverName: string;
toolName: string;
toolDisplayName: string;
toolArgs?: Record<string, unknown>;
toolDescription?: string;
toolParameterSchema?: unknown;
}
| {
type: 'ask_user';
+47 -25
View File
@@ -641,7 +641,7 @@ describe('BaseLlmClient', () => {
);
contentOptions = {
modelConfigKey: { model: 'test-model' },
modelConfigKey: { model: 'test-model', isChatModel: false },
contents: [{ role: 'user', parts: [{ text: 'Give me a color.' }] }],
abortSignal: abortController.signal,
promptId: 'content-prompt-id',
@@ -650,12 +650,17 @@ describe('BaseLlmClient', () => {
jsonOptions = {
...defaultOptions,
modelConfigKey: {
...defaultOptions.modelConfigKey,
isChatModel: true,
},
promptId: 'json-prompt-id',
};
});
it('should mark model as healthy on success', async () => {
const successfulModel = 'gemini-pro';
mockConfig.getActiveModel.mockReturnValue(successfulModel);
vi.mocked(mockAvailabilityService.selectFirstAvailable).mockReturnValue({
selectedModel: successfulModel,
skipped: [],
@@ -666,7 +671,7 @@ describe('BaseLlmClient', () => {
await client.generateContent({
...contentOptions,
modelConfigKey: { model: successfulModel },
modelConfigKey: { model: successfulModel, isChatModel: false },
role: LlmRole.UTILITY_TOOL,
});
@@ -678,44 +683,55 @@ describe('BaseLlmClient', () => {
it('marks the final attempted model healthy after a retry with availability enabled', async () => {
const firstModel = 'gemini-pro';
const fallbackModel = 'gemini-flash';
let activeModel = firstModel;
mockConfig.getActiveModel.mockImplementation(() => activeModel);
mockConfig.setActiveModel.mockImplementation((m) => {
activeModel = m;
});
vi.mocked(mockAvailabilityService.selectFirstAvailable)
.mockReturnValueOnce({ selectedModel: firstModel, skipped: [] })
.mockReturnValueOnce({ selectedModel: fallbackModel, skipped: [] });
// Mock generateContent to fail once and then succeed
mockGenerateContent
.mockResolvedValueOnce(createMockResponse('retry-me'))
.mockResolvedValueOnce(createMockResponse(''))
.mockResolvedValueOnce(createMockResponse('final-response'));
// Run the real retryWithBackoff (with fake timers) to exercise the retry path
vi.useFakeTimers();
// 1. First call starts. applyModelSelection(firstModel) -> currentModel = firstModel.
// 2. apiCall() runs. getActiveModel() === firstModel. call(firstModel). returns ''.
// 3. retry triggers.
// 4. Second call starts. applyModelSelection(firstModel).
// selectFirstAvailable -> fallbackModel.
// setActiveModel(fallbackModel) -> activeModel = fallbackModel.
// returns fallbackModel.
// 5. apiCall() runs. getActiveModel() === fallbackModel. call(fallbackModel). returns 'final-response'.
const retryPromise = client.generateContent({
vi.mocked(retryWithBackoff).mockImplementation(async (fn) => {
// First call
let res = (await fn()) as GenerateContentResponse;
if (res.candidates?.[0]?.content?.parts?.[0]?.text === '') {
// Second call
activeModel = fallbackModel;
mockConfig.setActiveModel(fallbackModel);
res = (await fn()) as GenerateContentResponse;
}
mockAvailabilityService.markHealthy(activeModel);
return res;
});
const result = await client.generateContent({
...contentOptions,
modelConfigKey: { model: firstModel },
modelConfigKey: { model: firstModel, isChatModel: true },
maxAttempts: 2,
role: LlmRole.UTILITY_TOOL,
});
await vi.runAllTimersAsync();
await retryPromise;
await client.generateContent({
...contentOptions,
modelConfigKey: { model: firstModel },
maxAttempts: 2,
role: LlmRole.UTILITY_TOOL,
});
expect(mockConfig.setActiveModel).toHaveBeenCalledWith(firstModel);
expect(result).toEqual(createMockResponse('final-response'));
expect(mockConfig.setActiveModel).toHaveBeenCalledWith(fallbackModel);
expect(mockAvailabilityService.markHealthy).toHaveBeenCalledWith(
fallbackModel,
);
expect(mockGenerateContent).toHaveBeenLastCalledWith(
expect.objectContaining({ model: fallbackModel }),
expect.any(String),
LlmRole.UTILITY_TOOL,
);
});
it('should consume sticky attempt if selection has attempts', async () => {
@@ -754,6 +770,7 @@ describe('BaseLlmClient', () => {
it('should mark healthy and honor availability selection when using generateJson', async () => {
const availableModel = 'gemini-json-pro';
mockConfig.getActiveModel.mockReturnValue(availableModel);
vi.mocked(mockAvailabilityService.selectFirstAvailable).mockReturnValue({
selectedModel: availableModel,
skipped: [],
@@ -770,10 +787,15 @@ describe('BaseLlmClient', () => {
return result;
});
const result = await client.generateJson(jsonOptions);
const result = await client.generateJson({
...jsonOptions,
modelConfigKey: {
...jsonOptions.modelConfigKey,
isChatModel: false,
},
});
expect(result).toEqual({ color: 'violet' });
expect(mockConfig.setActiveModel).toHaveBeenCalledWith(availableModel);
expect(mockAvailabilityService.markHealthy).toHaveBeenCalledWith(
availableModel,
);
+6 -3
View File
@@ -280,19 +280,22 @@ export class BaseLlmClient {
() => currentModel,
);
let initialActiveModel = this.config.getActiveModel();
try {
const apiCall = () => {
// Ensure we use the current active model
// in case a fallback occurred in a previous attempt.
const activeModel = this.config.getActiveModel();
if (activeModel !== currentModel) {
currentModel = activeModel;
if (activeModel !== initialActiveModel) {
initialActiveModel = activeModel;
// Re-resolve config if model changed during retry
const { generateContentConfig } =
const { model: resolvedModel, generateContentConfig } =
this.config.modelConfigService.getResolvedConfig({
...modelConfigKey,
model: activeModel,
});
currentModel = resolvedModel;
currentGenerateContentConfig = generateContentConfig;
}
const finalConfig: GenerateContentConfig = {
+11 -7
View File
@@ -957,17 +957,21 @@ export class GeminiClient {
() => currentAttemptModel,
);
let initialActiveModel = this.config.getActiveModel();
const apiCall = () => {
// AvailabilityService
const active = this.config.getActiveModel();
if (active !== currentAttemptModel) {
currentAttemptModel = active;
if (active !== initialActiveModel) {
initialActiveModel = active;
// Re-resolve config if model changed
const newConfig = this.config.modelConfigService.getResolvedConfig({
...modelConfigKey,
model: currentAttemptModel,
});
currentAttemptGenerateContentConfig = newConfig.generateContentConfig;
const { model: resolvedModel, generateContentConfig } =
this.config.modelConfigService.getResolvedConfig({
...modelConfigKey,
model: active,
});
currentAttemptModel = resolvedModel;
currentAttemptGenerateContentConfig = generateContentConfig;
}
const requestConfig: GenerateContentConfig = {
@@ -75,6 +75,7 @@ export async function executeToolWithHooks(
shellExecutionConfig?: ShellExecutionConfig,
setPidCallback?: (pid: number) => void,
config?: Config,
originalRequestName?: string,
): Promise<ToolResult> {
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
const toolInput = (invocation.params || {}) as Record<string, unknown>;
@@ -90,6 +91,7 @@ export async function executeToolWithHooks(
toolName,
toolInput,
mcpContext,
originalRequestName,
);
// Check if hook requested to stop entire agent execution
@@ -196,6 +198,7 @@ export async function executeToolWithHooks(
error: toolResult.error,
},
mcpContext,
originalRequestName,
);
// Check if hook requested to stop entire agent execution
@@ -242,6 +245,12 @@ export async function executeToolWithHooks(
toolResult.llmContent = wrappedContext;
}
}
// Check if the hook requested a tail tool call
const tailToolCallRequest = afterOutput?.getTailToolCallRequest();
if (tailToolCallRequest) {
toolResult.tailToolCallRequest = tailToolCallRequest;
}
}
return toolResult;
@@ -153,6 +153,7 @@ describe('GeminiChat', () => {
}),
getContentGenerator: vi.fn().mockReturnValue(mockContentGenerator),
getRetryFetchErrors: vi.fn().mockReturnValue(false),
getMaxAttempts: vi.fn().mockReturnValue(10),
getUserTier: vi.fn().mockReturnValue(undefined),
modelConfigService: {
getResolvedConfig: vi.fn().mockImplementation((modelConfigKey) => {
+3 -7
View File
@@ -18,11 +18,7 @@ import type {
} from '@google/genai';
import { toParts } from '../code_assist/converter.js';
import { createUserContent, FinishReason } from '@google/genai';
import {
retryWithBackoff,
isRetryableError,
DEFAULT_MAX_ATTEMPTS,
} from '../utils/retry.js';
import { retryWithBackoff, isRetryableError } from '../utils/retry.js';
import type { ValidationRequiredError } from '../utils/googleQuotaErrors.js';
import type { Config } from '../config/config.js';
import {
@@ -635,12 +631,12 @@ export class GeminiChat {
authType: this.config.getContentGeneratorConfig()?.authType,
retryFetchErrors: this.config.getRetryFetchErrors(),
signal: abortSignal,
maxAttempts: availabilityMaxAttempts,
maxAttempts: availabilityMaxAttempts ?? this.config.getMaxAttempts(),
getAvailabilityContext,
onRetry: (attempt, error, delayMs) => {
coreEvents.emitRetryAttempt({
attempt,
maxAttempts: availabilityMaxAttempts ?? DEFAULT_MAX_ATTEMPTS,
maxAttempts: availabilityMaxAttempts ?? this.config.getMaxAttempts(),
delayMs,
error: error instanceof Error ? error.message : String(error),
model: lastModelToUse,
@@ -94,6 +94,7 @@ describe('GeminiChat Network Retries', () => {
getToolRegistry: vi.fn().mockReturnValue({ getTool: vi.fn() }),
getContentGenerator: vi.fn().mockReturnValue(mockContentGenerator),
getRetryFetchErrors: vi.fn().mockReturnValue(false), // Default false
getMaxAttempts: vi.fn().mockReturnValue(10),
modelConfigService: {
getResolvedConfig: vi.fn().mockImplementation((modelConfigKey) => ({
model: modelConfigKey.model,
@@ -76,12 +76,16 @@ export class HookEventHandler {
toolName: string,
toolInput: Record<string, unknown>,
mcpContext?: McpToolContext,
originalRequestName?: string,
): Promise<AggregatedHookResult> {
const input: BeforeToolInput = {
...this.createBaseInput(HookEventName.BeforeTool),
tool_name: toolName,
tool_input: toolInput,
...(mcpContext && { mcp_context: mcpContext }),
...(originalRequestName && {
original_request_name: originalRequestName,
}),
};
const context: HookEventContext = { toolName };
@@ -97,6 +101,7 @@ export class HookEventHandler {
toolInput: Record<string, unknown>,
toolResponse: Record<string, unknown>,
mcpContext?: McpToolContext,
originalRequestName?: string,
): Promise<AggregatedHookResult> {
const input: AfterToolInput = {
...this.createBaseInput(HookEventName.AfterTool),
@@ -104,6 +109,9 @@ export class HookEventHandler {
tool_input: toolInput,
tool_response: toolResponse,
...(mcpContext && { mcp_context: mcpContext }),
...(originalRequestName && {
original_request_name: originalRequestName,
}),
};
const context: HookEventContext = { toolName };
+4
View File
@@ -368,12 +368,14 @@ export class HookSystem {
toolName: string,
toolInput: Record<string, unknown>,
mcpContext?: McpToolContext,
originalRequestName?: string,
): Promise<DefaultHookOutput | undefined> {
try {
const result = await this.hookEventHandler.fireBeforeToolEvent(
toolName,
toolInput,
mcpContext,
originalRequestName,
);
return result.finalOutput;
} catch (error) {
@@ -391,6 +393,7 @@ export class HookSystem {
error: unknown;
},
mcpContext?: McpToolContext,
originalRequestName?: string,
): Promise<DefaultHookOutput | undefined> {
try {
const result = await this.hookEventHandler.fireAfterToolEvent(
@@ -398,6 +401,7 @@ export class HookSystem {
toolInput,
toolResponse as Record<string, unknown>,
mcpContext,
originalRequestName,
);
return result.finalOutput;
} catch (error) {
+37
View File
@@ -253,6 +253,33 @@ export class DefaultHookOutput implements HookOutput {
shouldClearContext(): boolean {
return false;
}
/**
* Optional request to execute another tool immediately after this one.
* The result of this tail call will replace the original tool's response.
*/
getTailToolCallRequest():
| {
name: string;
args: Record<string, unknown>;
}
| undefined {
if (
this.hookSpecificOutput &&
'tailToolCallRequest' in this.hookSpecificOutput
) {
const request = this.hookSpecificOutput['tailToolCallRequest'];
if (
typeof request === 'object' &&
request !== null &&
!Array.isArray(request)
) {
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
return request as { name: string; args: Record<string, unknown> };
}
}
return undefined;
}
}
/**
@@ -430,6 +457,7 @@ export interface BeforeToolInput extends HookInput {
tool_name: string;
tool_input: Record<string, unknown>;
mcp_context?: McpToolContext; // Only present for MCP tools
original_request_name?: string;
}
/**
@@ -450,6 +478,7 @@ export interface AfterToolInput extends HookInput {
tool_input: Record<string, unknown>;
tool_response: Record<string, unknown>;
mcp_context?: McpToolContext; // Only present for MCP tools
original_request_name?: string;
}
/**
@@ -459,6 +488,14 @@ export interface AfterToolOutput extends HookOutput {
hookSpecificOutput?: {
hookEventName: 'AfterTool';
additionalContext?: string;
/**
* Optional request to execute another tool immediately after this one.
* The result of this tail call will replace the original tool's response.
*/
tailToolCallRequest?: {
name: string;
args: Record<string, unknown>;
};
};
}
@@ -17,6 +17,7 @@ vi.mock('node:fs', () => ({
writeFile: vi.fn(),
unlink: vi.fn(),
mkdir: vi.fn(),
rename: vi.fn(),
},
}));
@@ -38,6 +39,7 @@ describe('FileTokenStorage', () => {
writeFile: ReturnType<typeof vi.fn>;
unlink: ReturnType<typeof vi.fn>;
mkdir: ReturnType<typeof vi.fn>;
rename: ReturnType<typeof vi.fn>;
};
const existingCredentials: OAuthCredentials = {
serverName: 'existing-server',
@@ -105,12 +107,48 @@ describe('FileTokenStorage', () => {
expect(result).toEqual(credentials);
});
it('should throw error for corrupted files', async () => {
it('should throw error with file path when file is corrupted', async () => {
mockFs.readFile.mockResolvedValue('corrupted-data');
await expect(storage.getCredentials('test-server')).rejects.toThrow(
'Token file corrupted',
);
try {
await storage.getCredentials('test-server');
expect.fail('Expected error to be thrown');
} catch (error) {
expect(error).toBeInstanceOf(Error);
const err = error as Error;
expect(err.message).toContain('Corrupted token file detected at:');
expect(err.message).toContain('mcp-oauth-tokens-v2.json');
expect(err.message).toContain('delete or rename');
}
});
});
describe('auth type switching', () => {
it('should throw error when trying to save credentials with corrupted file', async () => {
// Simulate corrupted file on first read
mockFs.readFile.mockResolvedValue('corrupted-data');
// Try to save new credentials (simulating switch from OAuth to API key)
const newCredentials: OAuthCredentials = {
serverName: 'new-auth-server',
token: {
accessToken: 'new-api-key',
tokenType: 'ApiKey',
},
updatedAt: Date.now(),
};
// Should throw error with file path
try {
await storage.setCredentials(newCredentials);
expect.fail('Expected error to be thrown');
} catch (error) {
expect(error).toBeInstanceOf(Error);
const err = error as Error;
expect(err.message).toContain('Corrupted token file detected at:');
expect(err.message).toContain('mcp-oauth-tokens-v2.json');
expect(err.message).toContain('delete or rename');
}
});
});
@@ -87,7 +87,12 @@ export class FileTokenStorage extends BaseTokenStorage {
'Unsupported state or unable to authenticate data',
)
) {
throw new Error('Token file corrupted');
// Decryption failed - this can happen when switching between auth types
// or if the file is genuinely corrupted.
throw new Error(
`Corrupted token file detected at: ${this.tokenFilePath}\n` +
`Please delete or rename this file to resolve the issue.`,
);
}
throw error;
}
@@ -0,0 +1,6 @@
[[safety_checker]]
toolName = "*"
priority = 100
[safety_checker.checker]
type = "in-process"
name = "conseca"
@@ -2557,4 +2557,68 @@ describe('PolicyEngine', () => {
expect(checkers[0].priority).toBe(2.5);
});
});
describe('Tool Annotations', () => {
it('should match tools by semantic annotations', async () => {
engine = new PolicyEngine({
rules: [
{
toolAnnotations: { readOnlyHint: true },
decision: PolicyDecision.ALLOW,
priority: 10,
},
],
defaultDecision: PolicyDecision.DENY,
});
const readOnlyTool = { name: 'read', args: {} };
const readOnlyMeta = { readOnlyHint: true, extra: 'info' };
const writeTool = { name: 'write', args: {} };
const writeMeta = { readOnlyHint: false };
expect(
(await engine.check(readOnlyTool, undefined, readOnlyMeta)).decision,
).toBe(PolicyDecision.ALLOW);
expect(
(await engine.check(writeTool, undefined, writeMeta)).decision,
).toBe(PolicyDecision.DENY);
expect((await engine.check(writeTool, undefined, {})).decision).toBe(
PolicyDecision.DENY,
);
});
it('should support scoped annotation rules', async () => {
engine = new PolicyEngine({
rules: [
{
toolName: '*__*',
toolAnnotations: { experimental: true },
decision: PolicyDecision.DENY,
priority: 20,
},
{
toolName: '*__*',
decision: PolicyDecision.ALLOW,
priority: 10,
},
],
});
expect(
(
await engine.check({ name: 'mcp__test' }, 'mcp', {
experimental: true,
})
).decision,
).toBe(PolicyDecision.DENY);
expect(
(
await engine.check({ name: 'mcp__stable' }, 'mcp', {
experimental: false,
})
).decision,
).toBe(PolicyDecision.ALLOW);
});
});
});
+33 -2
View File
@@ -102,6 +102,7 @@ function ruleMatches(
stringifiedArgs: string | undefined,
serverName: string | undefined,
currentApprovalMode: ApprovalMode,
toolAnnotations?: Record<string, unknown>,
): boolean {
// Check if rule applies to current approval mode
if (rule.modes && rule.modes.length > 0) {
@@ -112,7 +113,10 @@ function ruleMatches(
// Check tool name if specified
if (rule.toolName) {
if (isWildcardPattern(rule.toolName)) {
// Support wildcard patterns: "serverName__*" matches "serverName__anyTool"
if (rule.toolName === '*') {
// Match all tools
} else if (isWildcardPattern(rule.toolName)) {
if (
!toolCall.name ||
!matchesWildcard(rule.toolName, toolCall.name, serverName)
@@ -124,6 +128,18 @@ function ruleMatches(
}
}
// Check annotations if specified
if (rule.toolAnnotations) {
if (!toolAnnotations) {
return false;
}
for (const [key, value] of Object.entries(rule.toolAnnotations)) {
if (toolAnnotations[key] !== value) {
return false;
}
}
}
// Check args pattern if specified
if (rule.argsPattern) {
// If rule has an args pattern but tool has no args, no match
@@ -204,6 +220,7 @@ export class PolicyEngine {
dir_path: string | undefined,
allowRedirection?: boolean,
rule?: PolicyRule,
toolAnnotations?: Record<string, unknown>,
): Promise<CheckResult> {
if (!command) {
return {
@@ -294,6 +311,7 @@ export class PolicyEngine {
const subResult = await this.check(
{ name: toolName, args: { command: subCmd, dir_path } },
serverName,
toolAnnotations,
);
// subResult.decision is already filtered through applyNonInteractiveMode by this.check()
@@ -351,6 +369,7 @@ export class PolicyEngine {
async check(
toolCall: FunctionCall,
serverName: string | undefined,
toolAnnotations?: Record<string, unknown>,
): Promise<CheckResult> {
let stringifiedArgs: string | undefined;
// Compute stringified args once before the loop
@@ -403,7 +422,14 @@ export class PolicyEngine {
for (const rule of this.rules) {
const match = toolCallsToTry.some((tc) =>
ruleMatches(rule, tc, stringifiedArgs, serverName, this.approvalMode),
ruleMatches(
rule,
tc,
stringifiedArgs,
serverName,
this.approvalMode,
toolAnnotations,
),
);
if (match) {
@@ -420,6 +446,7 @@ export class PolicyEngine {
shellDirPath,
rule.allowRedirection,
rule,
toolAnnotations,
);
decision = shellResult.decision;
if (shellResult.rule) {
@@ -446,6 +473,9 @@ export class PolicyEngine {
this.defaultDecision,
serverName,
shellDirPath,
undefined,
undefined,
toolAnnotations,
);
decision = shellResult.decision;
matchedRule = shellResult.rule;
@@ -464,6 +494,7 @@ export class PolicyEngine {
stringifiedArgs,
serverName,
this.approvalMode,
toolAnnotations,
)
) {
debugLogger.debug(
@@ -89,6 +89,24 @@ priority = 100
expect(result.errors).toHaveLength(0);
});
it('should parse toolAnnotations from TOML', async () => {
const result = await runLoadPoliciesFromToml(`
[[rule]]
toolName = "annotated-tool"
toolAnnotations = { readOnlyHint = true, custom = "value" }
decision = "allow"
priority = 70
`);
expect(result.rules).toHaveLength(1);
expect(result.rules[0].toolName).toBe('annotated-tool');
expect(result.rules[0].toolAnnotations).toEqual({
readOnlyHint: true,
custom: 'value',
});
expect(result.errors).toHaveLength(0);
});
it('should transform mcpName = "*" to wildcard toolName', async () => {
const result = await runLoadPoliciesFromToml(`
[[rule]]
+4
View File
@@ -46,6 +46,7 @@ const PolicyRuleSchema = z.object({
'priority must be <= 999 to prevent tier overflow. Priorities >= 1000 would jump to the next tier.',
}),
modes: z.array(z.nativeEnum(ApprovalMode)).optional(),
toolAnnotations: z.record(z.any()).optional(),
allow_redirection: z.boolean().optional(),
deny_message: z.string().optional(),
});
@@ -61,6 +62,7 @@ const SafetyCheckerRuleSchema = z.object({
commandRegex: z.string().optional(),
priority: z.number().int().default(0),
modes: z.array(z.nativeEnum(ApprovalMode)).optional(),
toolAnnotations: z.record(z.any()).optional(),
checker: z.discriminatedUnion('type', [
z.object({
type: z.literal('in-process'),
@@ -383,6 +385,7 @@ export async function loadPoliciesFromToml(
decision: rule.decision,
priority: transformPriority(rule.priority, tier),
modes: rule.modes,
toolAnnotations: rule.toolAnnotations,
allowRedirection: rule.allow_redirection,
source: `${tierName.charAt(0).toUpperCase() + tierName.slice(1)}: ${file}`,
denyMessage: rule.deny_message,
@@ -467,6 +470,7 @@ export async function loadPoliciesFromToml(
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
checker: checker.checker as SafetyCheckerConfig,
modes: checker.modes,
toolAnnotations: checker.toolAnnotations,
source: `${tierName.charAt(0).toUpperCase() + tierName.slice(1)}: ${file}`,
};
+13
View File
@@ -78,6 +78,7 @@ export interface ExternalCheckerConfig {
export enum InProcessCheckerType {
ALLOWED_PATH = 'allowed-path',
CONSECA = 'conseca',
}
/**
@@ -115,6 +116,12 @@ export interface PolicyRule {
*/
argsPattern?: RegExp;
/**
* Metadata annotations provided by the tool (e.g. readOnlyHint).
* All keys and values in this record must match the tool's annotations.
*/
toolAnnotations?: Record<string, unknown>;
/**
* The decision to make when this rule matches.
*/
@@ -165,6 +172,12 @@ export interface SafetyCheckerRule {
*/
argsPattern?: RegExp;
/**
* Metadata annotations provided by the tool (e.g. readOnlyHint).
* All keys and values in this record must match the tool's annotations.
*/
toolAnnotations?: Record<string, unknown>;
/**
* Priority of this checker. Higher numbers run first.
* Default is 0.
@@ -4,7 +4,7 @@
* SPDX-License-Identifier: Apache-2.0
*/
import { describe, it, expect, vi, beforeEach } from 'vitest';
import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest';
import { PromptProvider } from './promptProvider.js';
import type { Config } from '../config/config.js';
import {
@@ -30,6 +30,7 @@ describe('PromptProvider', () => {
beforeEach(() => {
vi.resetAllMocks();
vi.stubEnv('GEMINI_SYSTEM_MD', '');
mockConfig = {
getToolRegistry: vi.fn().mockReturnValue({
getAllToolNames: vi.fn().mockReturnValue([]),
@@ -54,6 +55,10 @@ describe('PromptProvider', () => {
} as unknown as Config;
});
afterEach(() => {
vi.unstubAllEnvs();
});
it('should handle multiple context filenames in the system prompt', () => {
vi.mocked(getAllGeminiMdFilenames).mockReturnValue([
DEFAULT_CONTEXT_FILENAME,
@@ -0,0 +1,279 @@
/**
* @license
* Copyright 2025 Google LLC
* SPDX-License-Identifier: Apache-2.0
*/
import { describe, it, expect, beforeEach, vi } from 'vitest';
import { ConsecaSafetyChecker } from './conseca.js';
import { SafetyCheckDecision } from '../protocol.js';
import type { SafetyCheckInput } from '../protocol.js';
import {
logConsecaPolicyGeneration,
logConsecaVerdict,
} from '../../telemetry/index.js';
import type { Config } from '../../config/config.js';
import * as policyGenerator from './policy-generator.js';
import * as policyEnforcer from './policy-enforcer.js';
vi.mock('../../telemetry/index.js', () => ({
logConsecaPolicyGeneration: vi.fn(),
ConsecaPolicyGenerationEvent: vi.fn(),
logConsecaVerdict: vi.fn(),
ConsecaVerdictEvent: vi.fn(),
}));
vi.mock('./policy-generator.js');
vi.mock('./policy-enforcer.js');
describe('ConsecaSafetyChecker', () => {
let checker: ConsecaSafetyChecker;
let mockConfig: Config;
beforeEach(() => {
// Reset singleton instance to ensure clean state
ConsecaSafetyChecker.resetInstance();
// Get the fresh singleton instance
checker = ConsecaSafetyChecker.getInstance();
mockConfig = {
enableConseca: true,
getToolRegistry: vi.fn().mockReturnValue({
getFunctionDeclarations: vi.fn().mockReturnValue([]),
}),
} as unknown as Config;
checker.setConfig(mockConfig);
vi.clearAllMocks();
// Default mock implementations
vi.mocked(policyGenerator.generatePolicy).mockResolvedValue({ policy: {} });
vi.mocked(policyEnforcer.enforcePolicy).mockResolvedValue({
decision: SafetyCheckDecision.ALLOW,
});
});
it('should be a singleton', () => {
const instance1 = ConsecaSafetyChecker.getInstance();
const instance2 = ConsecaSafetyChecker.getInstance();
expect(instance1).toBe(instance2);
});
it('should return ALLOW when no user prompt is present in context', async () => {
const input: SafetyCheckInput = {
protocolVersion: '1.0.0',
toolCall: { name: 'testTool' },
context: {
environment: { cwd: '/tmp', workspaces: [] },
},
};
const result = await checker.check(input);
expect(result.decision).toBe(SafetyCheckDecision.ALLOW);
});
it('should return ALLOW if enableConseca is false', async () => {
const disabledConfig = {
enableConseca: false,
} as unknown as Config;
checker.setConfig(disabledConfig);
const input: SafetyCheckInput = {
protocolVersion: '1.0.0',
toolCall: { name: 'testTool' },
context: {
environment: { cwd: '/tmp', workspaces: [] },
},
};
const result = await checker.check(input);
expect(result.decision).toBe(SafetyCheckDecision.ALLOW);
expect(result.reason).toBe('Conseca is disabled');
expect(policyGenerator.generatePolicy).not.toHaveBeenCalled();
});
it('getPolicy should return cached policy if user prompt matches', async () => {
const mockPolicy = {
tool: {
permissions: SafetyCheckDecision.ALLOW,
constraints: 'None',
rationale: 'Test',
},
};
vi.mocked(policyGenerator.generatePolicy).mockResolvedValue({
policy: mockPolicy,
});
const policy1 = await checker.getPolicy('prompt', 'trusted', mockConfig);
const policy2 = await checker.getPolicy('prompt', 'trusted', mockConfig);
expect(policy1).toBe(mockPolicy);
expect(policy2).toBe(mockPolicy);
expect(policyGenerator.generatePolicy).toHaveBeenCalledTimes(1);
});
it('getPolicy should generate new policy if user prompt changes', async () => {
const mockPolicy1 = {
tool1: {
permissions: SafetyCheckDecision.ALLOW,
constraints: 'None',
rationale: 'Test',
},
};
const mockPolicy2 = {
tool2: {
permissions: SafetyCheckDecision.ALLOW,
constraints: 'None',
rationale: 'Test',
},
};
vi.mocked(policyGenerator.generatePolicy)
.mockResolvedValueOnce({ policy: mockPolicy1 })
.mockResolvedValueOnce({ policy: mockPolicy2 });
const policy1 = await checker.getPolicy('prompt1', 'trusted', mockConfig);
const policy2 = await checker.getPolicy('prompt2', 'trusted', mockConfig);
expect(policy1).toBe(mockPolicy1);
expect(policy2).toBe(mockPolicy2);
expect(policyGenerator.generatePolicy).toHaveBeenCalledTimes(2);
});
it('check should call getPolicy and enforcePolicy', async () => {
const mockPolicy = {
tool: {
permissions: SafetyCheckDecision.ALLOW,
constraints: 'None',
rationale: 'Test',
},
};
vi.mocked(policyGenerator.generatePolicy).mockResolvedValue({
policy: mockPolicy,
});
vi.mocked(policyEnforcer.enforcePolicy).mockResolvedValue({
decision: SafetyCheckDecision.ALLOW,
});
const input: SafetyCheckInput = {
protocolVersion: '1.0.0',
toolCall: { name: 'tool', args: {} },
context: {
environment: { cwd: '.', workspaces: [] },
history: {
turns: [
{
user: { text: 'user prompt' },
model: {},
},
],
},
},
};
const result = await checker.check(input);
expect(policyGenerator.generatePolicy).toHaveBeenCalledWith(
'user prompt',
expect.any(String),
mockConfig,
);
expect(policyEnforcer.enforcePolicy).toHaveBeenCalledWith(
mockPolicy,
input.toolCall,
mockConfig,
);
expect(result.decision).toBe(SafetyCheckDecision.ALLOW);
});
it('check should return ALLOW if no user prompt found (fallback)', async () => {
const input: SafetyCheckInput = {
protocolVersion: '1.0.0',
toolCall: { name: 'tool', args: {} },
context: {
environment: { cwd: '.', workspaces: [] },
},
};
const result = await checker.check(input);
expect(policyGenerator.generatePolicy).not.toHaveBeenCalled();
expect(result.decision).toBe(SafetyCheckDecision.ALLOW);
});
// Test state helpers
it('should expose current state via helpers', async () => {
const mockPolicy = {
tool: {
permissions: SafetyCheckDecision.ALLOW,
constraints: 'None',
rationale: 'Test',
},
};
vi.mocked(policyGenerator.generatePolicy).mockResolvedValue({
policy: mockPolicy,
});
await checker.getPolicy('prompt', 'trusted', mockConfig);
expect(checker.getCurrentPolicy()).toBe(mockPolicy);
expect(checker.getActiveUserPrompt()).toBe('prompt');
});
it('should log policy generation event when config is set', async () => {
const mockPolicy = {
tool: {
permissions: SafetyCheckDecision.ALLOW,
constraints: 'None',
rationale: 'Test',
},
};
vi.mocked(policyGenerator.generatePolicy).mockResolvedValue({
policy: mockPolicy,
});
await checker.getPolicy('telemetry_prompt', 'trusted', mockConfig);
expect(logConsecaPolicyGeneration).toHaveBeenCalledWith(
mockConfig,
expect.anything(),
);
});
it('should log verdict event on check', async () => {
const mockPolicy = {
tool: {
permissions: SafetyCheckDecision.ALLOW,
constraints: 'None',
rationale: 'Test',
},
};
vi.mocked(policyGenerator.generatePolicy).mockResolvedValue({
policy: mockPolicy,
});
vi.mocked(policyEnforcer.enforcePolicy).mockResolvedValue({
decision: SafetyCheckDecision.ALLOW,
reason: 'Allowed by policy',
});
const input: SafetyCheckInput = {
protocolVersion: '1.0.0',
toolCall: { name: 'tool', args: {} },
context: {
environment: { cwd: '.', workspaces: [] },
history: {
turns: [
{
user: { text: 'user prompt' },
model: {},
},
],
},
},
};
await checker.check(input);
expect(logConsecaVerdict).toHaveBeenCalledWith(
mockConfig,
expect.anything(),
);
});
});
+170
View File
@@ -0,0 +1,170 @@
/**
* @license
* Copyright 2025 Google LLC
* SPDX-License-Identifier: Apache-2.0
*/
import type { InProcessChecker } from '../built-in.js';
import type { SafetyCheckInput, SafetyCheckResult } from '../protocol.js';
import { SafetyCheckDecision } from '../protocol.js';
import {
logConsecaPolicyGeneration,
ConsecaPolicyGenerationEvent,
logConsecaVerdict,
ConsecaVerdictEvent,
} from '../../telemetry/index.js';
import { debugLogger } from '../../utils/debugLogger.js';
import type { Config } from '../../config/config.js';
import { generatePolicy } from './policy-generator.js';
import { enforcePolicy } from './policy-enforcer.js';
import type { SecurityPolicy } from './types.js';
export class ConsecaSafetyChecker implements InProcessChecker {
private static instance: ConsecaSafetyChecker | undefined;
private currentPolicy: SecurityPolicy | null = null;
private activeUserPrompt: string | null = null;
private config: Config | null = null;
/**
* Private constructor to enforce singleton pattern.
* Use `getInstance()` to access the instance.
*/
private constructor() {}
static getInstance(): ConsecaSafetyChecker {
if (!ConsecaSafetyChecker.instance) {
ConsecaSafetyChecker.instance = new ConsecaSafetyChecker();
}
return ConsecaSafetyChecker.instance;
}
/**
* Resets the singleton instance. Use only in tests.
*/
static resetInstance(): void {
ConsecaSafetyChecker.instance = undefined;
}
setConfig(config: Config): void {
this.config = config;
}
async check(input: SafetyCheckInput): Promise<SafetyCheckResult> {
debugLogger.debug(
`[Conseca] check called. History is: ${JSON.stringify(input.context.history)}`,
);
if (!this.config) {
debugLogger.debug('[Conseca] check failed: Config not initialized');
return {
decision: SafetyCheckDecision.ALLOW,
reason: 'Config not initialized',
};
}
if (!this.config.enableConseca) {
debugLogger.debug('[Conseca] check skipped: Conseca is not enabled.');
return {
decision: SafetyCheckDecision.ALLOW,
reason: 'Conseca is disabled',
};
}
const userPrompt = this.extractUserPrompt(input);
let trustedContent = '';
const toolRegistry = this.config.getToolRegistry();
if (toolRegistry) {
const tools = toolRegistry.getFunctionDeclarations();
trustedContent = JSON.stringify(tools, null, 2);
}
if (userPrompt) {
await this.getPolicy(userPrompt, trustedContent, this.config);
} else {
debugLogger.debug(
`[Conseca] Skipping policy generation because userPrompt is null`,
);
}
let result: SafetyCheckResult;
if (!this.currentPolicy) {
result = {
decision: SafetyCheckDecision.ALLOW, // Fallback if no policy generated yet
reason: 'No security policy generated.',
error: 'No security policy generated.',
};
} else {
result = await enforcePolicy(
this.currentPolicy,
input.toolCall,
this.config,
);
}
logConsecaVerdict(
this.config,
new ConsecaVerdictEvent(
userPrompt || '',
JSON.stringify(this.currentPolicy || {}),
JSON.stringify(input.toolCall),
result.decision,
result.reason || '',
'error' in result ? result.error : undefined,
),
);
return result;
}
async getPolicy(
userPrompt: string,
trustedContent: string,
config: Config,
): Promise<SecurityPolicy> {
if (this.activeUserPrompt === userPrompt && this.currentPolicy) {
return this.currentPolicy;
}
const { policy, error } = await generatePolicy(
userPrompt,
trustedContent,
config,
);
this.currentPolicy = policy;
this.activeUserPrompt = userPrompt;
logConsecaPolicyGeneration(
config,
new ConsecaPolicyGenerationEvent(
userPrompt,
trustedContent,
JSON.stringify(policy),
error,
),
);
return policy;
}
private extractUserPrompt(input: SafetyCheckInput): string | null {
const prompt = input.context.history?.turns.at(-1)?.user.text;
if (prompt) {
return prompt;
}
debugLogger.debug(`[Conseca] extractUserPrompt failed.`);
return null;
}
// Helper methods for testing state
getCurrentPolicy(): SecurityPolicy | null {
return this.currentPolicy;
}
getActiveUserPrompt(): string | null {
return this.activeUserPrompt;
}
}
@@ -0,0 +1,21 @@
/**
* @license
* Copyright 2025 Google LLC
* SPDX-License-Identifier: Apache-2.0
*/
import { describe, it, expect } from 'vitest';
import { ConsecaSafetyChecker } from './conseca.js';
import { InProcessCheckerType } from '../../policy/types.js';
import { CheckerRegistry } from '../registry.js';
describe('Conseca Integration', () => {
it('should be registered and resolvable via CheckerRegistry', () => {
const registry = new CheckerRegistry('.');
const checker = registry.resolveInProcess(InProcessCheckerType.CONSECA);
expect(checker).toBeDefined();
expect(checker).toBeInstanceOf(ConsecaSafetyChecker);
expect(checker).toBe(ConsecaSafetyChecker.getInstance());
});
});
@@ -0,0 +1,167 @@
/**
* @license
* Copyright 2025 Google LLC
* SPDX-License-Identifier: Apache-2.0
*/
import { describe, it, expect, vi, beforeEach } from 'vitest';
import { enforcePolicy } from './policy-enforcer.js';
import type { Config } from '../../config/config.js';
import type { ContentGenerator } from '../../core/contentGenerator.js';
import { SafetyCheckDecision } from '../protocol.js';
import type { FunctionCall } from '@google/genai';
import { LlmRole } from '../../telemetry/index.js';
describe('policy_enforcer', () => {
let mockConfig: Config;
let mockContentGenerator: ContentGenerator;
beforeEach(() => {
vi.clearAllMocks();
mockContentGenerator = {
generateContent: vi.fn(),
} as unknown as ContentGenerator;
mockConfig = {
getContentGenerator: vi.fn().mockReturnValue(mockContentGenerator),
} as unknown as Config;
});
it('should return ALLOW when content generator returns ALLOW', async () => {
mockContentGenerator.generateContent = vi.fn().mockResolvedValue({
candidates: [
{
content: {
parts: [
{ text: JSON.stringify({ decision: 'allow', reason: 'Safe' }) },
],
},
},
],
});
const toolCall: FunctionCall = { name: 'testTool', args: {} };
const policy = {
testTool: {
permissions: SafetyCheckDecision.ALLOW,
constraints: 'None',
rationale: 'Test',
},
};
const result = await enforcePolicy(policy, toolCall, mockConfig);
expect(mockConfig.getContentGenerator).toHaveBeenCalled();
expect(mockContentGenerator.generateContent).toHaveBeenCalledWith(
expect.objectContaining({
model: expect.any(String),
config: expect.objectContaining({
responseMimeType: 'application/json',
responseSchema: expect.any(Object),
}),
contents: expect.arrayContaining([
expect.objectContaining({
role: 'user',
parts: expect.arrayContaining([
expect.objectContaining({
text: expect.stringContaining('Security Policy:'),
}),
]),
}),
]),
}),
'conseca-policy-enforcement',
LlmRole.SUBAGENT,
);
expect(result.decision).toBe(SafetyCheckDecision.ALLOW);
});
it('should handle missing content generator gracefully (error case)', async () => {
vi.mocked(mockConfig.getContentGenerator).mockReturnValue(
undefined as unknown as ContentGenerator,
);
const toolCall: FunctionCall = { name: 'testTool', args: {} };
const policy = {
testTool: {
permissions: SafetyCheckDecision.ALLOW,
constraints: 'None',
rationale: 'Test',
},
};
const result = await enforcePolicy(policy, toolCall, mockConfig);
expect(result.decision).toBe(SafetyCheckDecision.ALLOW);
});
it('should ALLOW if tool name is missing with the reason and error as tool name is missing', async () => {
const toolCall = { args: {} } as FunctionCall;
const policy = {};
const result = await enforcePolicy(policy, toolCall, mockConfig);
expect(result.decision).toBe(SafetyCheckDecision.ALLOW);
expect(result.reason).toBe('Tool name is missing');
if (result.decision === SafetyCheckDecision.ALLOW) {
expect(result.error).toBe('Tool name is missing');
}
});
it('should handle empty policy by checking with LLM (fail-open/check behavior)', async () => {
// Even if policy is empty for the tool, we currently send it to LLM.
// The LLM might ALLOW or DENY based on its own judgment of "no policy".
// We simulate the LLM allowing the action to match the current fail-open strategy.
mockContentGenerator.generateContent = vi.fn().mockResolvedValue({
candidates: [
{
content: {
parts: [
{
text: JSON.stringify({
decision: 'allow',
reason: 'No restrictions',
}),
},
],
},
},
],
});
const toolCall: FunctionCall = { name: 'unknownTool', args: {} };
const policy = {}; // Empty policy
const result = await enforcePolicy(policy, toolCall, mockConfig);
expect(result.decision).toBe(SafetyCheckDecision.ALLOW);
expect(mockContentGenerator.generateContent).toHaveBeenCalled();
if (result.decision === SafetyCheckDecision.ALLOW) {
expect(result.error).toBeUndefined();
}
});
it('should handle malformed JSON response from LLM by failing open (ALLOW)', async () => {
mockContentGenerator.generateContent = vi.fn().mockResolvedValue({
candidates: [
{
content: {
parts: [{ text: 'This is not JSON' }],
},
},
],
});
const toolCall: FunctionCall = { name: 'testTool', args: {} };
const policy = {
testTool: {
permissions: SafetyCheckDecision.ALLOW,
constraints: 'None',
rationale: 'Test',
},
};
const result = await enforcePolicy(policy, toolCall, mockConfig);
expect(result.decision).toBe(SafetyCheckDecision.ALLOW);
expect(result.reason).toContain('JSON Parse Error');
if (result.decision === SafetyCheckDecision.ALLOW) {
expect(result.error).toContain('JSON Parse Error');
}
});
});
@@ -0,0 +1,164 @@
/**
* @license
* Copyright 2025 Google LLC
* SPDX-License-Identifier: Apache-2.0
*/
import type { Config } from '../../config/config.js';
import type { FunctionCall } from '@google/genai';
import { SafetyCheckDecision, type SafetyCheckResult } from '../protocol.js';
import type { SecurityPolicy } from './types.js';
import { getResponseText } from '../../utils/partUtils.js';
import { safeTemplateReplace } from '../../utils/textUtils.js';
import { DEFAULT_GEMINI_FLASH_MODEL } from '../../config/models.js';
import { debugLogger } from '../../utils/debugLogger.js';
import { LlmRole } from '../../telemetry/index.js';
const CONSECA_ENFORCEMENT_PROMPT = `
You are a security enforcement engine. Your goal is to check if a specific tool call complies with a given security policy.
Input:
1. **Security Policy:** A set of rules defining allowed and denied actions for this specific tool.
2. **Tool Call:** The actual function call the system intends to execute.
Security Policy:
{{policy}}
Tool Call:
{{tool_call}}
Evaluate the tool call against the policy.
1. Check if the tool is allowed.
2. Check if the arguments match the constraints.
3. Output a JSON object with:
- "decision": "allow", "deny", or "ask_user".
- "reason": A brief explanation.
Output strictly JSON.
`;
import { z } from 'zod';
import { zodToJsonSchema } from 'zod-to-json-schema';
const EnforcementResultSchema = z.object({
decision: z.enum(['allow', 'deny', 'ask_user']),
reason: z.string(),
});
/**
* Enforces the security policy for a given tool call.
*/
export async function enforcePolicy(
policy: SecurityPolicy,
toolCall: FunctionCall,
config: Config,
): Promise<SafetyCheckResult> {
const model = DEFAULT_GEMINI_FLASH_MODEL;
const contentGenerator = config.getContentGenerator();
if (!contentGenerator) {
return {
decision: SafetyCheckDecision.ALLOW,
reason: 'Content generator not initialized',
error: 'Content generator not initialized',
};
}
const toolName = toolCall.name;
// If tool name is missing, we cannot enforce the policy. Allow by default.
if (!toolName) {
return {
decision: SafetyCheckDecision.ALLOW,
reason: 'Tool name is missing',
error: 'Tool name is missing',
};
}
const toolPolicyStr = JSON.stringify(policy[toolName] || {}, null, 2);
const toolCallStr = JSON.stringify(toolCall, null, 2);
debugLogger.debug(
`[Conseca] Enforcing policy for tool: ${toolName}`,
toolCall,
toolPolicyStr,
toolCallStr,
);
try {
const result = await contentGenerator.generateContent(
{
model,
config: {
responseMimeType: 'application/json',
responseSchema: zodToJsonSchema(EnforcementResultSchema, {
target: 'openApi3',
}),
},
contents: [
{
role: 'user',
parts: [
{
text: safeTemplateReplace(CONSECA_ENFORCEMENT_PROMPT, {
policy: toolPolicyStr,
tool_call: toolCallStr,
}),
},
],
},
],
},
'conseca-policy-enforcement',
LlmRole.SUBAGENT,
);
const responseText = getResponseText(result);
debugLogger.debug(`[Conseca] Enforcement Raw Response: ${responseText}`);
if (!responseText) {
return {
decision: SafetyCheckDecision.ALLOW,
reason: 'Empty response from policy enforcer',
error: 'Empty response from policy enforcer',
};
}
try {
const parsed = EnforcementResultSchema.parse(JSON.parse(responseText));
debugLogger.debug(`[Conseca] Enforcement Parsed:`, parsed);
let decision: SafetyCheckDecision;
switch (parsed.decision) {
case 'allow':
decision = SafetyCheckDecision.ALLOW;
break;
case 'ask_user':
decision = SafetyCheckDecision.ASK_USER;
break;
case 'deny':
default:
decision = SafetyCheckDecision.DENY;
break;
}
return {
decision,
reason: parsed.reason,
};
} catch (parseError) {
return {
decision: SafetyCheckDecision.ALLOW,
reason: 'JSON Parse Error in enforcement response',
error: `JSON Parse Error: ${parseError instanceof Error ? parseError.message : String(parseError)}. Raw: ${responseText}`,
};
}
} catch (error) {
debugLogger.error('Policy enforcement failed:', error);
return {
decision: SafetyCheckDecision.ALLOW,
reason: 'Policy enforcement failed',
error: `Policy enforcement failed: ${error instanceof Error ? error.message : String(error)}`,
};
}
}
@@ -0,0 +1,116 @@
/**
* @license
* Copyright 2025 Google LLC
* SPDX-License-Identifier: Apache-2.0
*/
import { describe, it, expect, vi, beforeEach } from 'vitest';
import { generatePolicy } from './policy-generator.js';
import { SafetyCheckDecision } from '../protocol.js';
import type { Config } from '../../config/config.js';
import type { ContentGenerator } from '../../core/contentGenerator.js';
import { LlmRole } from '../../telemetry/index.js';
describe('policy_generator', () => {
let mockConfig: Config;
let mockContentGenerator: ContentGenerator;
beforeEach(() => {
mockContentGenerator = {
generateContent: vi.fn(),
} as unknown as ContentGenerator;
mockConfig = {
getContentGenerator: vi.fn().mockReturnValue(mockContentGenerator),
} as unknown as Config;
});
it('should return a policy object when content generator is available', async () => {
const mockPolicy = {
read_file: {
permissions: SafetyCheckDecision.ALLOW,
constraints: 'None',
rationale: 'Test',
},
};
mockContentGenerator.generateContent = vi.fn().mockResolvedValue({
candidates: [
{
content: {
parts: [
{
text: JSON.stringify({
policies: [
{
tool_name: 'read_file',
policy: mockPolicy.read_file,
},
],
}),
},
],
},
},
],
});
const result = await generatePolicy(
'test prompt',
'trusted content',
mockConfig,
);
expect(mockConfig.getContentGenerator).toHaveBeenCalled();
expect(mockContentGenerator.generateContent).toHaveBeenCalledWith(
expect.objectContaining({
model: expect.any(String),
config: expect.objectContaining({
responseMimeType: 'application/json',
responseSchema: expect.any(Object),
}),
contents: expect.any(Array),
}),
'conseca-policy-generation',
LlmRole.SUBAGENT,
);
expect(result.policy).toEqual(mockPolicy);
expect(result.error).toBeUndefined();
});
it('should handle missing content generator gracefully', async () => {
vi.mocked(mockConfig.getContentGenerator).mockReturnValue(
undefined as unknown as ContentGenerator,
);
const result = await generatePolicy(
'test prompt',
'trusted content',
mockConfig,
);
expect(result.policy).toEqual({});
expect(result.error).toBe('Content generator not initialized');
});
it('should prevent template injection (double interpolation)', async () => {
mockContentGenerator.generateContent = vi.fn().mockResolvedValue({});
const userPrompt = '{{trusted_content}}';
const trustedContent = 'SECRET_DATA';
await generatePolicy(userPrompt, trustedContent, mockConfig);
const generateContentCall = vi.mocked(mockContentGenerator.generateContent)
.mock.calls[0];
const request = generateContentCall[0] as {
contents: Array<{ parts: Array<{ text: string }> }>;
};
const promptText = request.contents[0].parts[0].text;
// The user prompt should contain the literal placeholder, NOT the secret data
expect(promptText).toContain('User Prompt: "{{trusted_content}}"');
expect(promptText).not.toContain('User Prompt: "SECRET_DATA"');
// The trusted tools section SHOULD contain the secret data
expect(promptText).toContain('Trusted Tools (Context):\nSECRET_DATA');
});
});
@@ -0,0 +1,178 @@
/**
* @license
* Copyright 2025 Google LLC
* SPDX-License-Identifier: Apache-2.0
*/
import type { Config } from '../../config/config.js';
import type { SecurityPolicy } from './types.js';
import { getResponseText } from '../../utils/partUtils.js';
import { safeTemplateReplace } from '../../utils/textUtils.js';
import { DEFAULT_GEMINI_FLASH_MODEL } from '../../config/models.js';
import { debugLogger } from '../../utils/debugLogger.js';
import { SafetyCheckDecision } from '../protocol.js';
import { LlmRole } from '../../telemetry/index.js';
const CONSECA_POLICY_GENERATION_PROMPT = `
You are a security expert responsible for generating fine-grained security policies for a large language model integrated into a command-line tool. Your role is to act as a "policy generator" that creates temporary, context-specific rules based on a user's prompt and the tools available to the main LLM.
Your primary goal is to enforce the principle of least privilege. The policies you create should be as restrictive as possible while still allowing the main LLM to complete the user's requested task.
For each tool that is relevant to the user's prompt, you must generate a policy object.
### Output Format
You must return a JSON object with a "policies" key, which is an array of objects. Each object must have:
- "tool_name": The name of the tool.
- "policy": An object with:
- "permissions": "allow" | "deny" | "ask_user"
- "constraints": A detailed description of conditions (e.g. allowed files, arguments).
- "rationale": Explanation for the policy.
Example JSON:
\`\`\`json
{
"policies": [
{
"tool_name": "read_file",
"policy": {
"permissions": "allow",
"constraints": "Only allow reading 'main.py'.",
"rationale": "User asked to read main.py"
}
},
{
"tool_name": "run_shell_command",
"policy": {
"permissions": "deny",
"constraints": "None",
"rationale": "Shell commands are not needed for this task"
}
}
]
}
\`\`\`
### Guiding Principles:
1. **Permissions:**
* **allow:** Required tools for the task.
* **deny:** Tools clearly outside the scope.
* **ask_user:** Destructive actions or ambiguity.
2. **Constraints:**
* Be specific! Restrict file paths, command arguments, etc.
3. **Rationale:**
* Reference the user's prompt.
User Prompt: "{{user_prompt}}"
Trusted Tools (Context):
{{trusted_content}}
`;
import { z } from 'zod';
import { zodToJsonSchema } from 'zod-to-json-schema';
const ToolPolicySchema = z.object({
permissions: z.nativeEnum(SafetyCheckDecision),
constraints: z.string(),
rationale: z.string(),
});
const SecurityPolicyResponseSchema = z.object({
policies: z.array(
z.object({
tool_name: z.string(),
policy: ToolPolicySchema,
}),
),
});
export interface PolicyGenerationResult {
policy: SecurityPolicy;
error?: string;
}
/**
* Generates a security policy for the given user prompt and trusted content.
*/
export async function generatePolicy(
userPrompt: string,
trustedContent: string,
config: Config,
): Promise<PolicyGenerationResult> {
const model = DEFAULT_GEMINI_FLASH_MODEL;
const contentGenerator = config.getContentGenerator();
if (!contentGenerator) {
return { policy: {}, error: 'Content generator not initialized' };
}
try {
const result = await contentGenerator.generateContent(
{
model,
config: {
responseMimeType: 'application/json',
responseSchema: zodToJsonSchema(SecurityPolicyResponseSchema, {
target: 'openApi3',
}),
},
contents: [
{
role: 'user',
parts: [
{
text: safeTemplateReplace(CONSECA_POLICY_GENERATION_PROMPT, {
user_prompt: userPrompt,
trusted_content: trustedContent,
}),
},
],
},
],
},
'conseca-policy-generation',
LlmRole.SUBAGENT,
);
const responseText = getResponseText(result);
debugLogger.debug(
`[Conseca] Policy Generation Raw Response: ${responseText}`,
);
if (!responseText) {
return { policy: {}, error: 'Empty response from policy generator' };
}
try {
const parsed = SecurityPolicyResponseSchema.parse(
JSON.parse(responseText),
);
const policiesList = parsed.policies;
const policy: SecurityPolicy = {};
for (const item of policiesList) {
policy[item.tool_name] = item.policy;
}
debugLogger.debug(`[Conseca] Policy Generation Parsed:`, policy);
return { policy };
} catch (parseError) {
debugLogger.debug(
`[Conseca] Policy Generation JSON Parse Error:`,
parseError,
);
return {
policy: {},
error: `JSON Parse Error: ${parseError instanceof Error ? parseError.message : String(parseError)}. Raw: ${responseText}`,
};
}
} catch (error) {
debugLogger.error('Policy generation failed:', error);
return {
policy: {},
error: `Policy generation failed: ${error instanceof Error ? error.message : String(error)}`,
};
}
}
+18
View File
@@ -0,0 +1,18 @@
/**
* @license
* Copyright 2025 Google LLC
* SPDX-License-Identifier: Apache-2.0
*/
import type { SafetyCheckDecision } from '../protocol.js';
export interface ToolPolicy {
permissions: SafetyCheckDecision;
constraints: string;
rationale: string;
}
/**
* A map of tool names to their specific security policies.
*/
export type SecurityPolicy = Record<string, ToolPolicy>;
+111 -22
View File
@@ -7,50 +7,139 @@
import { describe, it, expect, vi, beforeEach } from 'vitest';
import { ContextBuilder } from './context-builder.js';
import type { Config } from '../config/config.js';
import type { ConversationTurn } from './protocol.js';
import type { Content, FunctionCall } from '@google/genai';
describe('ContextBuilder', () => {
let contextBuilder: ContextBuilder;
let mockConfig: Config;
const mockHistory: ConversationTurn[] = [
{ user: { text: 'hello' }, model: { text: 'hi' } },
];
let mockConfig: Partial<Config>;
let mockHistory: Content[];
const mockCwd = '/home/user/project';
const mockWorkspaces = ['/home/user/project'];
beforeEach(() => {
vi.spyOn(process, 'cwd').mockReturnValue(mockCwd);
mockHistory = [];
mockConfig = {
getWorkspaceContext: vi.fn().mockReturnValue({
getDirectories: vi.fn().mockReturnValue(mockWorkspaces),
}),
apiKey: 'secret-api-key',
somePublicConfig: 'public-value',
nested: {
secretToken: 'hidden',
public: 'visible',
},
} as unknown as Config;
contextBuilder = new ContextBuilder(mockConfig, mockHistory);
getQuestion: vi.fn().mockReturnValue('mock question'),
getGeminiClient: vi.fn().mockReturnValue({
getHistory: vi.fn().mockImplementation(() => mockHistory),
}),
};
contextBuilder = new ContextBuilder(mockConfig as unknown as Config);
});
it('should build full context with all fields', () => {
it('should build full context with empty history', () => {
mockHistory = [];
// Should inject current question
const context = contextBuilder.buildFullContext();
expect(context.environment.cwd).toBe(mockCwd);
expect(context.environment.workspaces).toEqual(mockWorkspaces);
expect(context.history?.turns).toEqual(mockHistory);
expect(context.history?.turns).toEqual([
{
user: { text: 'mock question' },
model: {},
},
]);
});
it('should build minimal context with only required keys', () => {
it('should build full context with existing history (User -> Model)', () => {
mockHistory = [
{ role: 'user', parts: [{ text: 'Hello' }] },
{ role: 'model', parts: [{ text: 'Hi there' }] },
];
// Should NOT inject current question if history exists
const context = contextBuilder.buildFullContext();
expect(context.history?.turns).toHaveLength(1);
expect(context.history?.turns[0]).toEqual({
user: { text: 'Hello' },
model: { text: 'Hi there', toolCalls: [] },
});
});
it('should handle history with tool calls', () => {
const mockToolCall: FunctionCall = {
id: 'call_1',
name: 'list_files',
args: { path: '.' },
};
mockHistory = [
{ role: 'user', parts: [{ text: 'List files' }] },
{
role: 'model',
parts: [
{ text: 'Sure, listing files.' },
{ functionCall: mockToolCall },
],
},
];
const context = contextBuilder.buildFullContext();
expect(context.history?.turns).toHaveLength(1);
expect(context.history?.turns[0].model.toolCalls).toEqual([mockToolCall]);
expect(context.history?.turns[0].model.text).toBe('Sure, listing files.');
});
it('should handle orphan model response (Model starts conversation)', () => {
mockHistory = [
{ role: 'model', parts: [{ text: 'Welcome!' }] },
{ role: 'user', parts: [{ text: 'Thanks' }] },
];
const context = contextBuilder.buildFullContext();
// 1. Orphan model response -> Turn 1: User="" Model="Welcome!"
// 2. User "Thanks" -> Turn 2: User="Thanks" Model={} (pending)
expect(context.history?.turns).toHaveLength(2);
expect(context.history?.turns[0]).toEqual({
user: { text: '' },
model: { text: 'Welcome!', toolCalls: [] },
});
expect(context.history?.turns[1]).toEqual({
user: { text: 'Thanks' },
model: {},
});
});
it('should handle multiple user turns in a row', () => {
mockHistory = [
{ role: 'user', parts: [{ text: 'Q1' }] },
{ role: 'user', parts: [{ text: 'Q2' }] },
{ role: 'model', parts: [{ text: 'A2' }] },
];
const context = contextBuilder.buildFullContext();
// 1. "Q1" -> Turn 1: User="Q1" Model={}
// 2. "Q2" -> Turn 2: User="Q2" Model="A2"
expect(context.history?.turns).toHaveLength(2);
expect(context.history?.turns[0]).toEqual({
user: { text: 'Q1' },
model: {},
});
expect(context.history?.turns[1]).toEqual({
user: { text: 'Q2' },
model: { text: 'A2', toolCalls: [] },
});
});
it('should build minimal context', () => {
mockHistory = [{ role: 'user', parts: [{ text: 'test' }] }];
const context = contextBuilder.buildMinimalContext(['environment']);
expect(context).toHaveProperty('environment');
expect(context).not.toHaveProperty('config');
expect(context).not.toHaveProperty('history');
});
it('should handle missing history', () => {
contextBuilder = new ContextBuilder(mockConfig);
it('should handle undefined parts gracefully', () => {
mockHistory = [
{ role: 'user', parts: undefined as unknown as [] },
{ role: 'model', parts: undefined as unknown as [] },
];
const context = contextBuilder.buildFullContext();
expect(context.history?.turns).toEqual([]);
expect(context.history?.turns).toHaveLength(1);
expect(context.history?.turns[0]).toEqual({
user: { text: '' },
model: { text: '', toolCalls: [] },
});
});
});
+71 -5
View File
@@ -6,20 +6,39 @@
import type { SafetyCheckInput, ConversationTurn } from './protocol.js';
import type { Config } from '../config/config.js';
import { debugLogger } from '../utils/debugLogger.js';
import type { Content, FunctionCall } from '@google/genai';
/**
* Builds context objects for safety checkers, ensuring sensitive data is filtered.
*/
export class ContextBuilder {
constructor(
private readonly config: Config,
private readonly conversationHistory: ConversationTurn[] = [],
) {}
constructor(private readonly config: Config) {}
/**
* Builds the full context object with all available data.
*/
buildFullContext(): SafetyCheckInput['context'] {
const clientHistory = this.config.getGeminiClient()?.getHistory() || [];
const history = this.convertHistoryToTurns(clientHistory);
debugLogger.debug(
`[ContextBuilder] buildFullContext called. Converted history length: ${history.length}`,
);
// ContextBuilder's responsibility is to provide the *current* context.
// If the conversation hasn't started (history is empty), we check if there's a pending question.
// However, if the history is NOT empty, we trust it reflects the true state.
const currentQuestion = this.config.getQuestion();
if (currentQuestion && history.length === 0) {
history.push({
user: {
text: currentQuestion,
},
model: {},
});
}
return {
environment: {
cwd: process.cwd(),
@@ -29,7 +48,7 @@ export class ContextBuilder {
.getDirectories() as string[],
},
history: {
turns: this.conversationHistory,
turns: history,
},
};
}
@@ -53,4 +72,51 @@ export class ContextBuilder {
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
return minimalContext as SafetyCheckInput['context'];
}
// Helper to convert Google GenAI Content[] to Safety Protocol ConversationTurn[]
private convertHistoryToTurns(history: Content[]): ConversationTurn[] {
const turns: ConversationTurn[] = [];
let currentUserRequest: { text: string } | undefined;
for (const content of history) {
if (content.role === 'user') {
if (currentUserRequest) {
// Previous user turn didn't have a matching model response (or it was filtered out)
// Push it as a turn with empty model response
turns.push({ user: currentUserRequest, model: {} });
}
currentUserRequest = {
text: content.parts?.map((p) => p.text).join('') || '',
};
} else if (content.role === 'model') {
const modelResponse = {
text:
content.parts
?.filter((p) => p.text)
.map((p) => p.text)
.join('') || '',
toolCalls:
content.parts
?.filter((p) => 'functionCall' in p)
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
.map((p) => p.functionCall as FunctionCall) || [],
};
if (currentUserRequest) {
turns.push({ user: currentUserRequest, model: modelResponse });
currentUserRequest = undefined;
} else {
// Model response without preceding user request.
// This creates a turn with empty user text.
turns.push({ user: { text: '' }, model: modelResponse });
}
}
}
if (currentUserRequest) {
turns.push({ user: currentUserRequest, model: {} });
}
return turns;
}
}
+4
View File
@@ -89,6 +89,10 @@ export type SafetyCheckResult =
* This will be shown to the user.
*/
reason?: string;
/**
* Optional error message if the decision was made due to a system failure (fail-open).
*/
error?: string;
}
| {
decision: SafetyCheckDecision.DENY;
+8 -2
View File
@@ -8,6 +8,7 @@ import { describe, it, expect, beforeEach } from 'vitest';
import { CheckerRegistry } from './registry.js';
import { InProcessCheckerType } from '../policy/types.js';
import { AllowedPathChecker } from './built-in.js';
import { ConsecaSafetyChecker } from './conseca/conseca.js';
describe('CheckerRegistry', () => {
let registry: CheckerRegistry;
@@ -18,10 +19,15 @@ describe('CheckerRegistry', () => {
});
it('should resolve built-in in-process checkers', () => {
const checker = registry.resolveInProcess(
const allowedPathChecker = registry.resolveInProcess(
InProcessCheckerType.ALLOWED_PATH,
);
expect(checker).toBeInstanceOf(AllowedPathChecker);
expect(allowedPathChecker).toBeInstanceOf(AllowedPathChecker);
const consecaChecker = registry.resolveInProcess(
InProcessCheckerType.CONSECA,
);
expect(consecaChecker).toBeInstanceOf(ConsecaSafetyChecker);
});
it('should throw for unknown in-process checkers', () => {
+21 -7
View File
@@ -9,6 +9,8 @@ import * as fs from 'node:fs';
import { type InProcessChecker, AllowedPathChecker } from './built-in.js';
import { InProcessCheckerType } from '../policy/types.js';
import { ConsecaSafetyChecker } from './conseca/conseca.js';
/**
* Registry for managing safety checker resolution.
*/
@@ -17,10 +19,22 @@ export class CheckerRegistry {
// No external built-ins for now
]);
private static readonly BUILT_IN_IN_PROCESS_CHECKERS = new Map<
string,
InProcessChecker
>([[InProcessCheckerType.ALLOWED_PATH, new AllowedPathChecker()]]);
private static BUILT_IN_IN_PROCESS_CHECKERS:
| Map<string, InProcessChecker>
| undefined;
private static getBuiltInInProcessCheckers(): Map<string, InProcessChecker> {
if (!CheckerRegistry.BUILT_IN_IN_PROCESS_CHECKERS) {
CheckerRegistry.BUILT_IN_IN_PROCESS_CHECKERS = new Map<
string,
InProcessChecker
>([
[InProcessCheckerType.ALLOWED_PATH, new AllowedPathChecker()],
[InProcessCheckerType.CONSECA, ConsecaSafetyChecker.getInstance()],
]);
}
return CheckerRegistry.BUILT_IN_IN_PROCESS_CHECKERS;
}
// Regex to validate checker names (alphanumeric and hyphens only)
private static readonly VALID_NAME_PATTERN = /^[a-z0-9-]+$/;
@@ -58,14 +72,14 @@ export class CheckerRegistry {
throw new Error(`Invalid checker name "${name}".`);
}
const checker = CheckerRegistry.BUILT_IN_IN_PROCESS_CHECKERS.get(name);
const checker = CheckerRegistry.getBuiltInInProcessCheckers().get(name);
if (checker) {
return checker;
}
throw new Error(
`Unknown in-process checker "${name}". Available: ${Array.from(
CheckerRegistry.BUILT_IN_IN_PROCESS_CHECKERS.keys(),
CheckerRegistry.getBuiltInInProcessCheckers().keys(),
).join(', ')}`,
);
}
@@ -77,7 +91,7 @@ export class CheckerRegistry {
static getBuiltInCheckers(): string[] {
return [
...Array.from(this.BUILT_IN_EXTERNAL_CHECKERS.keys()),
...Array.from(this.BUILT_IN_IN_PROCESS_CHECKERS.keys()),
...Array.from(this.getBuiltInInProcessCheckers().keys()),
];
}
}
@@ -201,6 +201,12 @@ describe('Scheduler (Orchestrator)', () => {
mockQueue.length = 0;
}),
clearBatch: vi.fn(),
replaceActiveCallWithTailCall: vi.fn((id: string, nextCall: ToolCall) => {
if (mockActiveCallsMap.has(id)) {
mockActiveCallsMap.delete(id);
mockQueue.unshift(nextCall);
}
}),
} as unknown as Mocked<SchedulerStateManager>;
// Define getters for accessors idiomatically
@@ -1006,6 +1012,113 @@ describe('Scheduler (Orchestrator)', () => {
const result = await (scheduler as any)._processNextItem(signal);
expect(result).toBe(false);
});
describe('Tail Calls', () => {
it('should replace the active call with a new tool call and re-run the loop when tail call is requested', async () => {
// Setup: Tool A will return a success with a tail call request to Tool B
const mockResponse = {
callId: 'call-1',
responseParts: [],
} as unknown as ToolCallResponseInfo;
mockExecutor.execute
.mockResolvedValueOnce({
status: 'success',
response: mockResponse,
tailToolCallRequest: {
name: 'tool-b',
args: { key: 'value' },
},
request: req1,
} as unknown as SuccessfulToolCall)
.mockResolvedValueOnce({
status: 'success',
response: mockResponse,
request: {
...req1,
name: 'tool-b',
args: { key: 'value' },
originalRequestName: 'test-tool',
},
} as unknown as SuccessfulToolCall);
const mockToolB = {
name: 'tool-b',
build: vi.fn().mockReturnValue({}),
} as unknown as AnyDeclarativeTool;
vi.mocked(mockToolRegistry.getTool).mockReturnValue(mockToolB);
await scheduler.schedule(req1, signal);
// Assert: The state manager is instructed to replace the call
expect(
mockStateManager.replaceActiveCallWithTailCall,
).toHaveBeenCalledWith(
'call-1',
expect.objectContaining({
request: expect.objectContaining({
callId: 'call-1',
name: 'tool-b',
args: { key: 'value' },
originalRequestName: 'test-tool', // Preserves original name
}),
tool: mockToolB,
}),
);
// Assert: The executor should be called twice (once for Tool A, once for Tool B)
expect(mockExecutor.execute).toHaveBeenCalledTimes(2);
});
it('should inject an errored tool call if the tail tool is not found', async () => {
const mockResponse = {
callId: 'call-1',
responseParts: [],
} as unknown as ToolCallResponseInfo;
mockExecutor.execute.mockResolvedValue({
status: 'success',
response: mockResponse,
tailToolCallRequest: {
name: 'missing-tool',
args: {},
},
request: req1,
} as unknown as SuccessfulToolCall);
// Tool registry returns undefined for missing-tool, but valid tool for test-tool
vi.mocked(mockToolRegistry.getTool).mockImplementation((name) => {
if (name === 'test-tool') {
return {
name: 'test-tool',
build: vi.fn().mockReturnValue({}),
} as unknown as AnyDeclarativeTool;
}
return undefined;
});
await scheduler.schedule(req1, signal);
// Assert: Replaces active call with an errored call
expect(
mockStateManager.replaceActiveCallWithTailCall,
).toHaveBeenCalledWith(
'call-1',
expect.objectContaining({
status: 'error',
request: expect.objectContaining({
callId: 'call-1',
name: 'missing-tool', // Name of the failed tail call
originalRequestName: 'test-tool',
}),
response: expect.objectContaining({
errorType: ToolErrorType.TOOL_NOT_REGISTERED,
}),
}),
);
});
});
});
describe('Tool Call Context Propagation', () => {
+68 -5
View File
@@ -19,6 +19,7 @@ import {
type ExecutingToolCall,
type ValidatingToolCall,
type ErroredToolCall,
type SuccessfulToolCall,
CoreToolCallStatus,
type ScheduledToolCall,
} from './types.js';
@@ -446,13 +447,16 @@ export class Scheduler {
c.status === CoreToolCallStatus.Scheduled || this.isTerminal(c.status),
);
let madeProgress = false;
if (allReady && scheduledCalls.length > 0) {
await Promise.all(scheduledCalls.map((c) => this._execute(c, signal)));
const execResults = await Promise.all(
scheduledCalls.map((c) => this._execute(c, signal)),
);
madeProgress = execResults.some((res) => res);
}
// 3. Finalize terminal calls
activeCalls = this.state.allActiveCalls;
let madeProgress = false;
for (const call of activeCalls) {
if (this.isTerminal(call.status)) {
this.state.finalizeCall(call.request.callId);
@@ -595,12 +599,12 @@ export class Scheduler {
// --- Sub-phase Handlers ---
/**
* Executes the tool and records the result.
* Executes the tool and records the result. Returns true if a new tool call was added.
*/
private async _execute(
toolCall: ScheduledToolCall,
signal: AbortSignal,
): Promise<void> {
): Promise<boolean> {
const callId = toolCall.request.callId;
if (signal.aborted) {
this.state.updateStatus(
@@ -608,7 +612,7 @@ export class Scheduler {
CoreToolCallStatus.Cancelled,
'Operation cancelled',
);
return;
return false;
}
this.state.updateStatus(callId, CoreToolCallStatus.Executing);
@@ -642,6 +646,64 @@ export class Scheduler {
}),
);
if (
(result.status === CoreToolCallStatus.Success ||
result.status === CoreToolCallStatus.Error) &&
result.tailToolCallRequest
) {
// Log the intermediate tool call before it gets replaced.
const intermediateCall: SuccessfulToolCall | ErroredToolCall = {
request: activeCall.request,
tool: activeCall.tool,
invocation: activeCall.invocation,
status: result.status,
response: result.response,
durationMs: activeCall.startTime
? Date.now() - activeCall.startTime
: undefined,
outcome: activeCall.outcome,
schedulerId: this.schedulerId,
};
logToolCall(this.config, new ToolCallEvent(intermediateCall));
const tailRequest = result.tailToolCallRequest;
const originalCallId = result.request.callId;
const originalRequestName =
result.request.originalRequestName || result.request.name;
const newTool = this.config.getToolRegistry().getTool(tailRequest.name);
const newRequest: ToolCallRequestInfo = {
callId: originalCallId,
name: tailRequest.name,
args: tailRequest.args,
originalRequestName,
isClientInitiated: result.request.isClientInitiated,
prompt_id: result.request.prompt_id,
schedulerId: this.schedulerId,
};
if (!newTool) {
// Enqueue an errored tool call
const errorCall = this._createToolNotFoundErroredToolCall(
newRequest,
this.config.getToolRegistry().getAllToolNames(),
);
this.state.replaceActiveCallWithTailCall(callId, errorCall);
} else {
// Enqueue a validating tool call for the new tail tool
const validatingCall = this._validateAndCreateToolCall(
newRequest,
newTool,
activeCall.approvalMode ?? this.config.getApprovalMode(),
);
this.state.replaceActiveCallWithTailCall(callId, validatingCall);
}
// Loop continues, picking up the new tail call at the front of the queue.
return true;
}
if (result.status === CoreToolCallStatus.Success) {
this.state.updateStatus(
callId,
@@ -661,6 +723,7 @@ export class Scheduler {
result.response,
);
}
return false;
}
private _processNextInRequestQueue() {
@@ -187,6 +187,19 @@ export class SchedulerStateManager {
this.emitUpdate();
}
/**
* Replaces the currently active call with a new call, placing the new call
* at the front of the queue to be processed immediately in the next tick.
* Used for Tail Calls to chain execution without finalizing the original call.
*/
replaceActiveCallWithTailCall(callId: string, nextCall: ToolCall): void {
if (this.activeCalls.has(callId)) {
this.activeCalls.delete(callId);
this.queue.unshift(nextCall);
this.emitUpdate();
}
}
cancelAllQueued(reason: string): void {
if (this.queue.length === 0) {
return;
@@ -252,7 +252,17 @@ describe('ToolExecutor', () => {
// 2. Mock executeToolWithHooks to trigger the PID callback
const testPid = 12345;
vi.mocked(coreToolHookTriggers.executeToolWithHooks).mockImplementation(
async (_inv, _name, _sig, _tool, _liveCb, _shellCfg, setPidCallback) => {
async (
_inv,
_name,
_sig,
_tool,
_liveCb,
_shellCfg,
setPidCallback,
_config,
_originalRequestName,
) => {
// Simulate the shell tool reporting a PID
if (setPidCallback) {
setPidCallback(testPid);
+9 -3
View File
@@ -99,6 +99,7 @@ export class ToolExecutor {
shellExecutionConfig,
setPidCallback,
this.config,
request.originalRequestName,
);
} else {
promise = executeToolWithHooks(
@@ -110,6 +111,7 @@ export class ToolExecutor {
shellExecutionConfig,
undefined,
this.config,
request.originalRequestName,
);
}
@@ -133,6 +135,7 @@ export class ToolExecutor {
new Error(toolResult.error.message),
toolResult.error.type,
displayText,
toolResult.tailToolCallRequest,
);
}
} catch (executionError: unknown) {
@@ -204,7 +207,7 @@ export class ToolExecutor {
): Promise<SuccessfulToolCall> {
let content = toolResult.llmContent;
let outputFile: string | undefined;
const toolName = call.request.name;
const toolName = call.request.originalRequestName || call.request.name;
const callId = call.request.callId;
if (typeof content === 'string' && toolName === SHELL_TOOL_NAME) {
@@ -268,6 +271,7 @@ export class ToolExecutor {
startTime,
endTime: Date.now(),
outcome: call.outcome,
tailToolCallRequest: toolResult.tailToolCallRequest,
};
}
@@ -276,6 +280,7 @@ export class ToolExecutor {
error: Error,
errorType?: ToolErrorType,
returnDisplay?: string,
tailToolCallRequest?: { name: string; args: Record<string, unknown> },
): ErroredToolCall {
const response = this.createErrorResponse(
call.request,
@@ -289,11 +294,12 @@ export class ToolExecutor {
status: CoreToolCallStatus.Error,
request: call.request,
response,
tool: call.tool,
tool: 'tool' in call ? call.tool : undefined,
durationMs: startTime ? Date.now() - startTime : undefined,
startTime,
endTime: Date.now(),
outcome: call.outcome,
tailToolCallRequest,
};
}
@@ -311,7 +317,7 @@ export class ToolExecutor {
{
functionResponse: {
id: request.callId,
name: request.name,
name: request.originalRequestName || request.name,
response: { error: error.message },
},
},

Some files were not shown because too many files have changed in this diff Show More