@rallycry/conveyor-agent 7.3.0 → 7.3.2

package/dist/{chunk-5MTPAVNZ.js → chunk-WDGSLU5S.js}

@@ -416,7 +416,14 @@ ${q.question}${q.options.length ? "\n" + q.options.map((o) => `- ${o.label}: ${o
     const apiUrl = this.config.apiUrl;
     if (!codespaceName || !apiUrl) return false;
     try {
-      const response = await fetch(`${apiUrl}/api/codespace/bootstrap/${codespaceName}`);
+      const bootstrapToken = process.env.CONVEYOR_BOOTSTRAP_TOKEN;
+      const headers = {};
+      if (bootstrapToken) {
+        headers["x-codespace-token"] = bootstrapToken;
+      }
+      const response = await fetch(`${apiUrl}/api/codespace/bootstrap/${codespaceName}`, {
+        headers
+      });
       if (!response.ok) return false;
       const config = await response.json();
       if (config.envVars?.CLAUDE_CODE_OAUTH_TOKEN) {
@@ -1522,6 +1529,13 @@ function buildDiscoveryPrompt(context, runnerMode) {
     `- Include testing requirements and acceptance criteria`,
     `- Set \`storyPointValue\` based on estimated complexity`,
     ``,
+    `### Plan Verification Requirements`,
+    `Every plan MUST include a **Testing / Verification** section enumerating:`,
+    `- The quality gates: \`bun run lint\`, \`bun run typecheck\`, \`bun run test\` (or scoped equivalents)`,
+    `- Any task-specific end-to-end checks (manual UI walk-through, API smoke test, migration dry-run, etc.)`,
+    `- For refactors: line-count / file-size expectations and a public-API-surface diff check`,
+    `This ensures executing agents and reviewers have a clear definition of "done" and prevents silent regressions.`,
+    ``,
     `### Completing Planning`,
     `Once ALL checklist items above are done, call the **ExitPlanMode** tool.`,
     `- Required before ExitPlanMode will succeed: **plan** (via update_task), **story points** (via update_task_properties), **title** (via update_task_properties)`,
@@ -1595,7 +1609,15 @@ function buildModePrompt(agentMode, context, runnerMode) {
           `- Goal: coordinate child task execution and ensure all children complete successfully`
         ] : [
           `- If this is a leaf task (no children): execute the plan directly`,
-          `- Goal: implement the plan, run tests, open a PR when done`
+          `- Goal: implement the plan, verify quality gates, open a PR when done`,
+          ``,
+          `### Pre-PR Verification Checklist`,
+          `Before calling \`mcp__conveyor__create_pull_request\`, verify ALL of the following pass:`,
+          `1. \`bun run lint\` \u2014 no new lint errors`,
+          `2. \`bun run typecheck\` \u2014 no new type errors`,
+          `3. \`bun run test\` \u2014 relevant suites pass (scope to the affected package if the full run is prohibitively slow, and state the scope in the PR description)`,
+          `If a gate fails, fix it before opening the PR. Do NOT open PRs with known failing gates.`,
+          `For refactors: also run \`git diff ${context?.baseBranch ?? "dev"}..HEAD\` and confirm the public API surface (exports, function signatures) has no unintended breaking changes.`
         ]
       ];
       return parts.join("\n");
@@ -2087,6 +2109,7 @@ CRITICAL: You are in Auto mode. Do NOT report status, ask for confirmation, or g
     `Your FIRST action should be reading the relevant source files mentioned in the plan, then writing code. Do NOT run install, build, lint, test, or dev server commands first \u2014 the environment is already set up.`,
     `Work on the git branch "${context.githubBranch}". Stay on this branch for the entire task. Do not checkout or create other branches.`,
     `Your replies are visible to the team in chat \u2014 briefly describe what you're doing when you begin meaningful implementation, and again when the PR is ready.`,
+    `Before opening the PR, run the quality gates: \`bun run lint && bun run typecheck && bun run test\`. Fix any failures. Do NOT open a PR with known failing gates \u2014 scope the test run to affected packages if the full suite is prohibitively slow, and state the scope in the PR body.`,
     `When finished, use the mcp__conveyor__create_pull_request tool to open a PR. Do NOT use gh CLI or any other method to create PRs.`
   ];
   if (isAutoMode) {
@@ -2143,6 +2166,7 @@ New messages since your last run:`,
     ...newMessages.map((m) => `[${m.userName ?? "user"}]: ${m.content}`),
     `
 Address the requested changes directly. Do NOT re-investigate the codebase from scratch or write a new plan \u2014 go straight to implementing the feedback.`,
+    `Before pushing or opening a PR, re-run the quality gates: \`bun run lint && bun run typecheck && bun run test\`. Fix any failures \u2014 relaunches are the most common place verification gets skipped.`,
     `Implement your updates and open a PR when finished.`
   ];
   if (context.githubPRUrl) {
@@ -2236,7 +2260,7 @@ import { z } from "zod";
 function buildReadTaskChatTool(connection) {
   return defineTool(
     "read_task_chat",
-    "Read recent messages from a task chat. Omit task_id to read the current task's chat, or provide a child task ID to read a child's chat.",
+    "Read recent human/user chat messages for a task. When to use: you need to see user instructions, questions, or feedback posted to the task chat. Omit task_id for the current task, or pass a child task ID to read a child's chat. When NOT to use: for agent reasoning, tool calls, or setup/dev-server output \u2014 use get_task_cli instead. Returns: JSON array of recent chat messages (default 20).",
     {
       limit: z.number().optional().describe("Number of recent messages to fetch (default 20)"),
       task_id: z.string().optional().describe("Child task ID to read chat from. Omit to read the current task's chat.")
@@ -2263,7 +2287,7 @@ function buildReadTaskChatTool(connection) {
 function buildGetTaskPlanTool(connection) {
   return defineTool(
     "get_task_plan",
-    "Re-read the latest task plan in case it was updated",
+    "Re-read the current task's plan. When to use: the user just said they updated the plan mid-session or explicitly asked you to re-read it. When NOT to use: to fetch general task info (title, status, description) \u2014 use get_task. The plan was already included in your initial context, so do not call this speculatively. Returns: the plan markdown string.",
     {},
     async () => {
       try {
@@ -2281,7 +2305,7 @@ function buildGetTaskPlanTool(connection) {
 function buildGetTaskTool(connection) {
   return defineTool(
     "get_task",
-    "Look up a task by slug or ID to get its title, description, plan, and status",
+    "Look up any task by slug or ID. When to use: you need the current title, description, plan, status, branch, PR info, or story points for this or a related task. When NOT to use: to list child tasks (use list_subtasks), or to re-read only the current plan (use get_task_plan). Returns: JSON task object including id, slug, title, description, plan, status, branch, githubPRNumber, githubPRUrl, storyPointValue.",
     {
       slug_or_id: z.string().describe("The task slug (e.g. 'my-task') or CUID")
     },
@@ -2304,7 +2328,7 @@ function buildGetTaskTool(connection) {
 function buildGetTaskCliTool(connection) {
   return defineTool(
     "get_task_cli",
-    "Read CLI execution logs from a task. Returns agent reasoning, tool calls, setup output, and other execution events. Use 'source' to filter: 'agent' for agent reasoning/tool calls only, 'application' for setup/dev-server output only.",
+    "Read CLI execution logs from a task \u2014 agent reasoning, tool calls, and setup/dev-server output. When to use: inspecting what another agent did, debugging build/setup output, or reviewing prior tool calls. Filter with source='agent' for reasoning and tool calls only, source='application' for setup/dev-server output only. When NOT to use: for human/user chat \u2014 use read_task_chat. Returns: newline-joined event log lines each prefixed with `[timestamp] [type]` and formatted content.",
     {
       task_id: z.string().optional().describe("Task ID or slug. Omit to read logs from the current task."),
       source: z.enum(["agent", "application"]).optional().describe("Filter by log source. Omit for all logs."),
@@ -2337,7 +2361,7 @@ function buildGetTaskCliTool(connection) {
 function buildListTaskFilesTool(connection) {
   return defineTool(
     "list_task_files",
-    "List all files attached to this task with metadata (name, type, size) and download URLs",
+    "List all files attached to this task with metadata. When to use: before fetching a specific file, to see what is available and how large each is. When NOT to use: to read a file's contents \u2014 use get_task_file. Returns: JSON array of file metadata (id, name, mimeType, size, downloadUrl), plus inline image blocks for any attached images.",
     {},
     async () => {
       try {
@@ -2367,7 +2391,7 @@ function buildListTaskFilesTool(connection) {
 function buildGetTaskFileTool(connection) {
   return defineTool(
     "get_task_file",
-    "Get a specific task file's content and download URL by file ID",
+    "Fetch one task file's content plus metadata by file ID. When to use: you have a file ID from list_task_files and need its content or download URL. Call list_task_files first to discover IDs and check sizes \u2014 large binaries may be truncated by the service's size limit. When NOT to use: to enumerate attachments \u2014 use list_task_files. Returns: JSON metadata; for images an inline image block is also returned, and text content is embedded inline when available.",
     { fileId: z.string().describe("The file ID to retrieve") },
     async ({ fileId }) => {
       try {
@@ -2410,7 +2434,7 @@ import { z as z2 } from "zod";
 function buildGetDependenciesTool(connection) {
   return defineTool(
     "get_dependencies",
-    "Get this task's dependencies and their current status (met = merged to dev)",
+    "Get this task's dependencies and their current met/unmet status. When to use: confirming that blockers have been merged before starting work, or investigating why a task cannot start. When NOT to use: to look up a specific task's state \u2014 use get_task. Returns: JSON array of dependency objects (slug, title, status, and whether the dependency is met \u2014 met means merged to dev).",
     {},
     async () => {
       try {
@@ -2430,15 +2454,19 @@ function buildGetDependenciesTool(connection) {
 function buildGetSuggestionsTool(connection) {
   return defineTool(
     "get_suggestions",
-    "List project suggestions sorted by vote score. Use this to see what the team thinks is important.",
+    "List project suggestions sorted by vote score. When to use: seeing what the team thinks is important, finding a suggestion to vote on, or checking for duplicates before create_suggestion. Filter by status (Planning, Open, InProgress, ReviewPR, ReviewDev, ReviewLive, Complete, Cancelled) or cap results with limit (default 20). When NOT to use: for task lists \u2014 suggestions are project-level idea records, not tasks. Returns: JSON array of suggestions with id, title, description, status, score.",
     {
-      status: z2.string().optional().describe("Filter by status: Open, Accepted, Rejected, Implemented"),
-      limit: z2.number().optional().describe("Max results (default 20)")
+      status: z2.string().optional().describe(
+        "Filter by status: Planning, Open, InProgress, ReviewPR, ReviewDev, ReviewLive, Complete, Cancelled"
+      ),
+      limit: z2.number().int().min(1).max(100).optional().describe("Max results (default 20)")
     },
-    async ({ status: _status, limit: _limit }) => {
+    async ({ status, limit }) => {
       try {
         const suggestions = await connection.call("getSuggestions", {
-          sessionId: connection.sessionId
+          sessionId: connection.sessionId,
+          status,
+          limit
         });
         if (suggestions.length === 0) {
           return textResult("No suggestions found.");
@@ -2459,7 +2487,7 @@ import { z as z3 } from "zod";
 function buildPostToChatTool(connection) {
   return defineTool(
     "post_to_chat",
-    "Post a message to a task chat. Your normal replies already appear in chat \u2014 only use this for explicit out-of-band updates or posting to a child task's chat.",
+    "Post an out-of-band message to a task chat. When to use: the user explicitly asked for a status update that must appear in chat, or you need to message a child task's chat (pass its ID as task_id). When NOT to use: for normal responses \u2014 your regular replies already appear in the current task's chat automatically. Returns: confirmation string.",
     {
       message: z3.string().describe("The message to post to the team"),
       task_id: z3.string().optional().describe("Child task ID to post to. Omit to post to the current task's chat.")
@@ -2487,7 +2515,7 @@ function buildPostToChatTool(connection) {
 function buildForceUpdateTaskStatusTool(connection) {
   return defineTool(
     "force_update_task_status",
-    "EMERGENCY ONLY: Force-override a task's Kanban status. Status transitions happen automatically (building sets InProgress, PR creation sets ReviewPR, merge sets ReviewDev). Only use this if an automatic transition failed or a task is stuck in the wrong status. Omit task_id to update the current task, or provide a child task ID.",
+    "EMERGENCY ONLY: force-override a task's Kanban status. When to use: an automatic transition failed and the task is wedged \u2014 e.g. a PR was merged but status did not advance to ReviewDev, a build completed but status is still InProgress with no session running, or a child is stuck in ReviewPR after its PR was closed. When NOT to use: in normal flow \u2014 building, PR creation, and merge each transition status automatically. Omit task_id for the current task; pass a child task ID to update a child. Returns: confirmation string.",
     {
       status: z3.enum(["InProgress", "ReviewPR", "ReviewDev", "Complete"]).describe("The new status for the task"),
       task_id: z3.string().optional().describe("Child task ID to update. Omit to update the current task.")
@@ -2519,7 +2547,7 @@ function buildForceUpdateTaskStatusTool(connection) {
 function buildCreatePullRequestTool(connection, config) {
   return defineTool(
     "create_pull_request",
-    "Create a GitHub pull request for this task. Automatically stages uncommitted changes, commits them, and pushes before creating the PR. Use this instead of gh CLI or git commands to create PRs.",
+    "Create a GitHub pull request for this task. Runs in order: 1) stage any uncommitted changes, 2) commit them (using commitMessage, or a default derived from title), 3) push to origin, 4) create the PR via GitHub API. When to use: opening the PR for this task \u2014 always use this instead of the gh CLI or raw git. Common failures: missing remote branch, a PR already open for this branch, or an expired git token (retry after the refresh). Returns: confirmation string with the new PR number and URL.",
     {
       title: z3.string().describe("The PR title"),
       body: z3.string().describe("The PR description/body in markdown"),
@@ -2605,7 +2633,7 @@ Troubleshooting:
 function buildAddDependencyTool(connection) {
   return defineTool(
     "add_dependency",
-    "Add a dependency \u2014 this task cannot start until the specified task is merged to dev",
+    "Add a blocking dependency \u2014 this task cannot start until the named task is merged to dev. When to use: you discovered work that must land before this task can proceed. When NOT to use: to track work that should happen after this task \u2014 use create_follow_up_task instead. Returns: confirmation string.",
     {
       depends_on_slug_or_id: z3.string().describe("Slug or ID of the task this task depends on")
     },
@@ -2627,7 +2655,7 @@ function buildAddDependencyTool(connection) {
 function buildRemoveDependencyTool(connection) {
   return defineTool(
     "remove_dependency",
-    "Remove a dependency from this task",
+    "Remove a previously added dependency from this task. When to use: the dependency was added in error or is no longer relevant. Returns: confirmation string.",
     {
       depends_on_slug_or_id: z3.string().describe("Slug or ID of the task to remove as dependency")
     },
@@ -2649,7 +2677,7 @@ function buildRemoveDependencyTool(connection) {
 function buildCreateFollowUpTaskTool(connection) {
   return defineTool(
     "create_follow_up_task",
-    "Create a follow-up task in this project that depends on the current task. Use for out-of-scope work, v1.1 features, or cleanup that should happen after this task merges.",
+    "Create a follow-up task that depends on the current task. When to use: out-of-scope work, v1.1 features, or cleanup that should happen after this task merges. The new task will be unblocked once the current task is merged. When NOT to use: to add a blocker to this task \u2014 use add_dependency instead. Returns: confirmation string with the new task's slug.",
     {
       title: z3.string().describe("Follow-up task title"),
       description: z3.string().optional().describe("Brief description of the follow-up work"),
@@ -2679,7 +2707,7 @@ function buildCreateFollowUpTaskTool(connection) {
 function buildCreateSuggestionTool(connection) {
   return defineTool(
     "create_suggestion",
-    "Suggest a feature, improvement, or idea for the project. If you want to recommend something \u2014 a document, a rule, a feature, a task, an optimization \u2014 use this tool. If a similar suggestion already exists, your vote will be added to it instead of creating a duplicate.",
+    "Suggest a feature, improvement, rule, or idea for the project. When to use: recommending anything project-scoped \u2014 a document, a rule, a feature, a task, an optimization. If a similar suggestion already exists the service will dedupe and record your upvote instead. When NOT to use: for actionable work scoped to the current task \u2014 open a follow-up task. Returns: confirmation string with the suggestion id (and mergedIntoId if deduped).",
     {
       title: z3.string().describe("Short title for the suggestion"),
       description: z3.string().optional().describe(
@@ -2712,7 +2740,7 @@ function buildCreateSuggestionTool(connection) {
 function buildVoteSuggestionTool(connection) {
   return defineTool(
     "vote_suggestion",
-    "Vote on a project suggestion. Use +1 to upvote or -1 to downvote.",
+    "Vote +1 or -1 on a project suggestion. When to use: expressing support or disagreement with a specific suggestion returned from get_suggestions. Returns: confirmation string with the suggestion's updated score.",
     {
       suggestion_id: z3.string().describe("The suggestion ID to vote on"),
       value: z3.number().refine((v) => v === 1 || v === -1, { message: "Value must be 1 or -1" }).describe("+1 to upvote, -1 to downvote")
@@ -2761,7 +2789,7 @@ var SP_DESCRIPTION = "Story point value (1=Common, 2=Magic, 3=Rare, 5=Unique)";
 function buildUpdateTaskTool(connection) {
   return defineTool(
     "update_task",
-    "Save the finalized task plan and/or description",
+    "Save the finalized plan and/or description to the current task. When to use: in Plan mode, after reaching alignment with the user on the approach. When NOT to use: for child/subtasks \u2014 use update_subtask. This tool does not change status or other properties. Returns: confirmation string.",
     {
       plan: z4.string().optional().describe("The task plan in markdown"),
       description: z4.string().optional().describe("Updated task description")
@@ -2783,7 +2811,7 @@ function buildUpdateTaskTool(connection) {
 function buildCreateSubtaskTool(connection) {
   return defineTool(
     "create_subtask",
-    "Create a subtask under the current parent task. Use for breaking complex tasks into smaller pieces.",
+    "Create a subtask under the current parent task. When to use: breaking a complex parent task into smaller pieces during planning. When NOT to use: for follow-ups that depend on this task but are not children \u2014 use create_follow_up_task. Returns: confirmation string with the new subtask id.",
     {
       title: z4.string().describe("Subtask title"),
       description: z4.string().optional().describe("Brief description"),
@@ -2813,7 +2841,7 @@ function buildCreateSubtaskTool(connection) {
 function buildUpdateSubtaskTool(connection) {
   return defineTool(
     "update_subtask",
-    "Update an existing subtask's fields",
+    "Update an existing subtask's fields (title, description, plan, ordinal, storyPointValue). When to use: refining a child's plan or reordering the subtask queue. When NOT to use: to change status (pack tools do that automatically) or to update the current task itself (use update_task). Returns: confirmation string.",
     {
       subtaskId: z4.string().describe("The subtask ID to update"),
       title: z4.string().optional(),
@@ -2842,7 +2870,7 @@ function buildUpdateSubtaskTool(connection) {
 function buildDeleteSubtaskTool(connection) {
   return defineTool(
     "delete_subtask",
-    "Delete a subtask",
+    "Delete a subtask by id. When to use: a subtask was created in error or is no longer needed. Returns: confirmation string.",
     { subtaskId: z4.string().describe("The subtask ID to delete") },
     async ({ subtaskId }) => {
       try {
@@ -2860,7 +2888,7 @@ function buildDeleteSubtaskTool(connection) {
 function buildListSubtasksTool(connection) {
   return defineTool(
     "list_subtasks",
-    "List all subtasks under the current parent task. Returns status, PR info (githubPRNumber, githubPRUrl), agent assignment (agentId), and plan for each child.",
+    "List all subtasks under the current parent task. When to use: coordinating child work \u2014 check who is running, which children are ready for review, or which are blocked. When NOT to use: for non-child related tasks \u2014 use get_task. Returns: JSON array with id, slug, title, status, storyPointValue, githubPRNumber, githubPRUrl, agentId, and plan for each child.",
     {},
     async () => {
       try {
@@ -2879,7 +2907,7 @@ function buildPackTools(connection) {
   return [
     defineTool(
       "start_child_cloud_build",
-      "Start a cloud build for a child task. The child must be in Open status with story points and an agent assigned.",
+      "Start a cloud build (codespace) for a child task. Preconditions: the child must be in Open status, have a story point value, and have an agent assigned. When NOT to use: if the child is already InProgress or past Open \u2014 it is already building or done. Returns: confirmation string including the child task id that the build was started for.",
       {
         childTaskId: z4.string().describe("The child task ID to start a cloud build for")
       },
@@ -2899,7 +2927,7 @@ function buildPackTools(connection) {
     ),
     defineTool(
       "stop_child_build",
-      "Stop a running cloud build for a child task. Sends a stop signal to the child agent.",
+      "Send a graceful stop signal to a running child build's agent. When to use: you decided the child should halt (plan changed, deadlock, scope pivot) and want the agent to exit cleanly. This is not a force-kill \u2014 the agent may take a moment to wind down. Returns: confirmation string.",
       {
         childTaskId: z4.string().describe("The child task ID whose build should be stopped")
       },
@@ -2919,7 +2947,7 @@ function buildPackTools(connection) {
     ),
     defineTool(
       "approve_and_merge_pr",
-      "Approve and merge a child task's pull request. Only succeeds if all CI/CD checks are passing. Returns an error if checks are pending (retry after waiting) or failed (investigate). The child task must be in ReviewPR status.",
+      "Approve and merge a child task's PR. Preconditions: child is in ReviewPR and you have reviewed the diff. When NOT to use: if review surfaced issues \u2014 post to the child's chat or escalate instead. Returns: { childTaskId, prNumber, merged }. merged=true means the merge happened and status is now ReviewDev. merged=false means the merge is queued for GitHub automerge (CI still running) \u2014 do NOT proceed as if merged; wait for the child's status to transition to ReviewDev. If checks have failed, this call errors \u2014 investigate the failure.",
       {
         childTaskId: z4.string().describe("The child task ID whose PR should be approved and merged")
       },
@@ -3007,7 +3035,7 @@ function buildCodeReviewTools(connection) {
   return [
     defineTool(
       "approve_code_review",
-      "Approve the code review. Use this when the code passes all review criteria and is ready to merge.",
+      "Approve the code review and exit. When to use: the diff passes all review criteria and is ready to merge. When NOT to use: if any substantive issue remains \u2014 call request_code_changes with a structured issues[] list. This tool takes only a summary; it does not accept an issues array (by design \u2014 an approval should have no blocking issues). Returns: confirmation string.",
       {
         summary: z6.string().describe("Brief summary of what was reviewed and why it looks good")
       },
@@ -3030,7 +3058,7 @@ ${summary}`;
     ),
     defineTool(
       "request_code_changes",
-      "Request changes during code review. Use this when substantive issues are found that need to be fixed before merge.",
+      "Request changes during code review and exit. When to use: substantive issues were found that must be fixed before merge. Each entry in issues[] is { file: string, line?: number, severity: 'critical' | 'major' | 'minor', description: string }. When NOT to use: for an approval \u2014 use approve_code_review. Returns: confirmation string.",
       {
         issues: z6.array(
           z6.object({
@@ -3624,7 +3652,7 @@ function buildClientInspectionTools(manager) {
   return [
     defineTool(
       "debug_inspect_client_paused",
-      "When the client-side debugger is paused at a breakpoint, returns the call stack and local variables. Includes React component state, props, and hooks when paused inside a component.",
+      "When the client-side (browser) debugger is paused at a breakpoint, returns the call stack and local variables from the paused frame. When to use: after a client breakpoint hits, to see what is in scope in the browser. When NOT to use: for server pauses \u2014 use debug_inspect_paused. Returns: JSON with side='client', reason, hitBreakpoints, callStack, and localVariables. If the debugger paused between turns and has since resumed, queued breakpoint hits are returned instead.",
       {},
       async () => {
         const clientOrErr = requirePlaywrightClient(manager);
@@ -3667,7 +3695,7 @@ ${JSON.stringify(queuedHits, null, 2)}`
     ),
     defineTool(
       "debug_evaluate_client",
-      "Evaluate a JavaScript expression in the client-side browser context. When paused at a client breakpoint, evaluates in the paused scope. Can access DOM, window, React internals, etc.",
+      "Evaluate a JavaScript expression in the browser context. When paused at a client breakpoint, runs in the paused scope (use frameIndex to pick a frame); otherwise runs in the page's global scope with access to DOM, window, and React internals. Side effects are real \u2014 navigations, DOM mutations, and function calls execute. Prefer read-only expressions unless you specifically want to mutate page state. Returns: `(type) value`.",
       {
         expression: z8.string().describe("JavaScript expression to evaluate in the browser context"),
         frameIndex: z8.number().optional().describe("Call stack frame index (0 = top frame). Defaults to the top frame.")
@@ -3741,7 +3769,7 @@ function buildClientInteractionTools(manager) {
     ),
     defineTool(
       "debug_navigate_client",
-      "Navigate the headless browser to a specific URL. Use this to reproduce specific flows or visit different pages.",
+      "Navigate the headless browser to a URL. When to use: reproducing a specific flow or visiting a different page. Waits for `domcontentloaded` before returning (Playwright's default ~30s navigation timeout applies). Returns: confirmation string with the resolved current URL.",
       {
         url: z8.string().describe("URL to navigate to (e.g., http://localhost:3000/dashboard)")
       },
@@ -3758,7 +3786,7 @@ function buildClientInteractionTools(manager) {
     ),
     defineTool(
       "debug_click_client",
-      "Click an element on the page in the headless browser. Use CSS selectors to target elements. Useful for reproducing bugs by interacting with the UI programmatically.",
+      "Click an element in the headless browser by CSS selector. When to use: reproducing bugs by interacting with the UI programmatically. Playwright auto-waits for the element to be visible, stable, and enabled before clicking, up to a 10s timeout \u2014 a miss throws and is reported in the failure message. Returns: confirmation string.",
       {
         selector: z8.string().describe(
           "CSS selector of the element to click (e.g., 'button.submit', '#login-form input[type=submit]')"
@@ -3931,7 +3959,7 @@ function buildDebugLifecycleTools(manager) {
   return [
     defineTool(
       "debug_enter_mode",
-      "Activate debug mode: restarts the dev server with Node.js --inspect flag and connects the CDP debugger. Optionally launch a headless Chromium browser for client-side debugging. Use serverSide for backend breakpoints, clientSide for frontend breakpoints, or both for full-stack.",
+      "Activate debug mode. When to use: at the start of a debug session before setting breakpoints or probes. Restarts the dev server with Node.js --inspect for server-side CDP, and/or launches a headless Chromium via Playwright for client-side. Default: if neither serverSide nor clientSide is passed, server-side only is activated. Pass clientSide=true for frontend debugging (previewUrl required), or set both for full-stack. Returns: activation status string listing active modes, an echoed hypothesis if given, and a warning if source maps are missing.",
       {
         hypothesis: z9.string().optional().describe("Your hypothesis about the bug \u2014 helps track debugging intent"),
         serverSide: z9.boolean().optional().describe(
@@ -4077,7 +4105,7 @@ ${JSON.stringify(queuedHits, null, 2)}`
     ),
     defineTool(
       "debug_evaluate",
-      "Evaluate a JavaScript expression in the current paused scope (or globally if not paused). When paused, use frameIndex to evaluate in a specific call frame.",
+      "Evaluate a JavaScript expression server-side in the Node process. When paused at a breakpoint the expression runs in that frame's scope (use frameIndex to pick a different frame); otherwise it runs in the global scope. Side effects are real \u2014 assignments, function calls, and I/O all execute. Prefer read-only expressions unless you specifically want to mutate state. Returns: `(type) value`.",
       {
         expression: z9.string().describe("The JavaScript expression to evaluate"),
         frameIndex: z9.number().optional().describe("Call stack frame index (0 = top frame). Defaults to the top frame.")
@@ -4176,7 +4204,7 @@ function buildProbeResultTools(manager) {
   return [
     defineTool(
       "debug_get_probe_results",
-      "Fetch captured probe hit data. Returns expression values from each time a probed line executed.",
+      "Fetch captured probe hit data. When to use: after triggering the code path for a probe set by debug_add_probe. Filtering: pass label to filter directly, or probeId to look up and filter by that probe's label; if both are passed, label wins and probeId is ignored. Returns: grouped text with per-probe hit count, timestamps, and captured expression values.",
       {
         probeId: z9.string().optional().describe("Filter results by probe ID (resolves to its label)"),
         label: z9.string().optional().describe("Filter results by probe label"),
@@ -5586,6 +5614,7 @@ var QueryBridge = class {
       }
     } finally {
       this.mode.pendingModeRestart = false;
+      this._abortController = null;
     }
   }
   // ── QueryHost construction ──────────────────────────────────────────
@@ -6691,8 +6720,8 @@ function buildTaskListTools(connection) {
       { annotations: { readOnlyHint: true } }
     ),
     defineTool(
-      "get_task",
-      "Get detailed information about a task including chat messages, child tasks, and session.",
+      "get_project_task",
+      "Get detailed information about a task in this project (chat messages, child tasks, session). Project-runner scope.",
       { task_id: z10.string().describe("The task ID to look up") },
       async ({ task_id }) => {
         try {
@@ -6790,8 +6819,8 @@ function buildMutationTools2(connection) {
       }
     ),
     defineTool(
-      "update_task",
-      "Update an existing task's title, description, plan, status, or assignee.",
+      "update_project_task",
+      "Update an existing task's title, description, plan, status, or assignee. Project-runner scope.",
       {
         task_id: z10.string().describe("The task ID to update"),
         title: z10.string().optional().describe("New title"),
@@ -7341,7 +7370,7 @@ function spawnChildAgent(assignment, workDir) {
       CONVEYOR_TASK_TOKEN: taskToken,
       CONVEYOR_TASK_ID: taskId,
       ...sessionId ? { CONVEYOR_SESSION_ID: sessionId } : {},
-      CONVEYOR_MODE: agentMode === "code-review" ? "code-review" : mode,
+      CONVEYOR_MODE: mode,
       CONVEYOR_WORKSPACE: workDir,
       CONVEYOR_USE_WORKTREE: "false",
       CONVEYOR_AGENT_MODE: effectiveAgentMode,
@@ -7395,7 +7424,7 @@ async function killAgent(agent, taskId) {
 async function handleAssignment(assignment, activeAgents, projectDir, connection) {
   const { taskId, mode } = assignment;
   const shortId = taskId.slice(0, 8);
-  const agentKey = assignment.agentMode === "code-review" ? `${taskId}:code-review` : taskId;
+  const agentKey = mode === "code-review" ? `${taskId}:code-review` : taskId;
   const existing = activeAgents.get(agentKey);
   if (existing) {
     if (existing.process.exitCode === null) {
@@ -7796,4 +7825,4 @@ export {
   loadForwardPorts,
   loadConveyorConfig
 };
-//# sourceMappingURL=chunk-5MTPAVNZ.js.map
+//# sourceMappingURL=chunk-WDGSLU5S.js.map