main.js · part 223

60 text occurrences from desktop/Cursor.app/Contents/Resources/app/extensions/cursor-agent-host/dist/main.js, part 223. Every entry preserves the shipped literal and its saved verdict or selection reason.

File contents and all parts · All files

Poll a background shell.

Poll a background ${t?"shell or subagent":"shell"} job. For work that does not have a ${t?"task id":"shell id"}, you can omit the ${t?"task_id":"shell_id"} arg to sleep for the full `block_until_ms` duration (prefer this over sleeping in the shell, because it renders nicely to the user).

Monitor backgrounded jobs as follows:
- Never poll a task whose tool result says it was "manually backgrounded by the user".
- When you spawn a command directly into the background (`block_until_ms: 0`), check status immediately by reading the output file to confirm the command didn't fail to start.
- Poll repeatedly to monitor by using this tool between checks (set `block_until_ms` to control how long to wait). If the file gets large, read from the end of the file to capture the latest content.
- Pick your polling intervals using best guess/judgment based on any knowledge you have about the command and its expected runtime, and any output from monitoring the job. When no new output, exponential backoff is a good strategy (e.g. 2000ms, 4000ms, 8000ms, 16000ms...), using educated guess for min and max wait.${r?"\n- Shell only guidance:\n  - Waiting until a regex matches the output can be useful for e.g. known startup/status/error logs.\n  - HARD STOPPING CONSTRAINT: Don't stop polling until (a) job terminates, (b) the command reaches a healthy steady state (only for non-terminating command, e.g. dev server/watcher), or (c) command is hung - follow guidance below.\n  - Output file header has `pid` and `running_for_ms` (updated every 5000ms).\n  - When finished, footer with `exit_code` and `elapsed_ms` appears (regex only matches the body, not header/footer).\n  - If taking longer than expected and the command seems like it is hung (use judgment based on type of command), kill the process if safe to do so using the pid that appears in the header. If possible, try to fix the hang and proceed.":""}

shell or subagent

task id

shell id

- Shell only guidance:
  - Waiting until a regex matches the output can be useful for e.g. known startup/status/error logs.
  - HARD STOPPING CONSTRAINT: Don't stop polling until (a) job terminates, (b) the command reaches a healthy steady state (only for non-terminating command, e.g. dev server/watcher), or (c) command is hung - follow guidance below.
  - Output file header has `pid` and `running_for_ms` (updated every 5000ms).
  - When finished, footer with `exit_code` and `elapsed_ms` appears (regex only matches the body, not header/footer).
  - If taking longer than expected and the command seems like it is hung (use judgment based on type of command), kill the process if safe to do so using the pid that appears in the header. If possible, try to fix the hang and proceed.

Unhandled version: ${t}

Must pass a task id or wait for a nonzero duration.

Regex awaiting is not available for subagents. Run Await without regex, or await a shell task id instead.

Awaiting subagent tasks requires a non-empty conversationId in the agent context

Awaiting subagent ids is not available in this environment (subagent executor resource is not registered).

Unknown legacy subagent await result

Unhandled legacy subagent result case: ${e}

Subagent not found

Failed to await subagent task

Unknown subagent await result

Unhandled subagent await result case: ${e}

Cannot await shell task: terminals folder is unavailable

Failed to read shell output

No shell found for id ${g}

Pattern matched: ${e.regexMatch??""}

Pattern did NOT match.

Slept briefly.

Slept for ${Math.max(1,Math.ceil(t/1e3))}s.

${t} Sleep released early: a new user message is arriving.

Task completed in ${(e.runtimeMs??BigInt(0)).toString()}ms with exit code: ${void 0!==r?String(r):"unknown"}.

Task complete.

${o} ${t}

${o}
output_file_path: ${e.outputFilePath}
output_length: ${e.outputLength.toString()}

Task still running after ${(e.runtimeMs??BigInt(0)).toString()}ms...

Task still running.

${r} Wait released early: a new user message is arriving.

${r} ${t}

${o}
output_file_path: ${e.outputFilePath}
output_length: ${e.outputLength.toString()}

Error awaiting task: ${s.value.error}

Unknown error

Object expected.

Symbol.asyncDispose is not defined.

Symbol.dispose is not defined.

Object not disposable.

An error was suppressed during disposal.

File paths to get AI attribution for. Use repository-relative git paths (for example: backend/server/src/app.ts), not absolute workspace paths.

Optional start line (1-indexed). If not provided, gets AI attribution for entire file. Applies to all files.

Optional end line (1-indexed). If not provided, gets AI attribution to end of file. Applies to all files.

Commit hashes to get AI attribution for. Returns AI conversation summaries for each commit.

Controls output shape and verbosity. Use "summary" for conversation-level rollups and "detailed" for per-commit/per-range output.

Optional cap on commits returned after filtering/de-duplication. Must be between 1 and 200.

When false, omit detailed line ranges from attribution output.

Either file_paths or commit_hashes must be provided

Unhandled version: ${e}

Trace how code was built with AI for files and commits.

Treat this as git blame for AI-built changes: it reveals which conversations produced the code, why it was implemented a certain way, what alternatives were considered, and the intent behind decisions.

When to use this tool:
- The user asks about a file, module, or area of code and wants to understand it deeply -- use this alongside reading the file to add context about how and why it was built.
- Questions about history, evolution, or recent changes to code (e.g. "tell me about this file", "what's changed here recently", "how has this evolved"). For concept-level questions without a specific file (e.g. "how has plan mode evolved?"), first find the relevant files, then use this tool on them.
- Questions about authorship or decision-making (e.g. "who wrote this", "why is it done this way", "what tradeoffs were made").
- Before planning changes to existing code, to understand the intent behind the current implementation and avoid undoing deliberate design decisions.
- When used during planning, surface deliberate design decisions, past reversions, and intentional tradeoffs as constraints. Explain how the proposed changes relate to those prior decisions.

Output includes conversation context depth:
- Conversation title / tldr / overview / summary bullets
- Models used across the conversation and commit-linked sections

Query shape policy:
- File-first (`file_paths`) when the user anchor is code location. Always pass repository-relative git paths (for example `backend/server/src/app.ts`), not absolute workspace paths like `/Users/.../projects/repo/backend/server/src/app.ts`.
- Commit-first (`commit_hashes`) when the user anchor is a change event or narrative.
- Use both when both anchors are known (high precision, less noise).
- Use `file_paths` to compare a bounded set of files; use `commit_hashes` for stack/series narrative.
- For `file_paths`, pass repository-relative git paths (for example, paths shown in `git diff`), not absolute local filesystem paths.

Output strategy:
- Start with `output_mode: "summary"` + modest `max_commits`.
- Pivot to `output_mode: "detailed"` with line ranges once a target section is identified.
- If file query is empty but activity is expected, pivot to commit query.
- If commit query is noisy, add file filters.

Transcript fallback:
- If transcript pointers/context are available and you need deeper decision narrative, inspect transcript snippets after this tool (read in chunks, do not read entire transcript files at once).

Returns commit history with conversation-linked lineage details suitable for code archaeology and implementation-intent analysis.

[ai-attribution] fetching attribution

[ai-attribution] attribution fetched successfully

[ai-attribution] failed to fetch attribution

Failed to fetch AI attribution data. This may be a temporary issue - please try again later.

AI attribution output

Failed to get AI attribution: ${e.result.value.error}

An unknown error occurred.

Failed to fetch AI attribution data. This may be a temporary issue - please try again later.

Canvas validation timed out; if this save targeted a canvas, re-save the file to confirm its status and path.