main.js · part 218

60 text occurrences from desktop/Cursor.app/Contents/Resources/app/extensions/cursor-agent-host/dist/main.js, part 218. Every entry preserves the shipped literal and its saved verdict or selection reason.

File contents and all parts · All files

Video attachments are only supported for Gemini models

Could not detect mime type for attachment: ${t}

Steer-driven subagent background handoff failed

Client-side subagent failed or was aborted

Unknown subagent error

Unknown subagent result

Client-side subagent completed successfully

WritableIterable is closed

The subagent's connection closed before it finished (the run was torn down mid-flight). This is usually transient — please try again.

Client-side subagent execution interrupted by stream teardown

Client-side subagent execution threw error

<system_reminder>
${c}
</system_reminder>

${i}

Creating subagent and starting execution

Subagent exceeded execution timeout of ${pe}ms

Spawn local and cloud subagents natively.

Failed to parse arguments

Invalid arguments:
argument: ${t}

${e.path.length>0?e.path.join("."):"argument"}: ${e.message}

Invalid arguments:
${e.join("\n")}

Object expected.

Symbol.asyncDispose is not defined.

Symbol.dispose is not defined.

Object not disposable.

An error was suppressed during disposal.

Failed to record initial checkpoint outcome

Time for initializeConversation (system prompt, rules, turn creation)

Time to resolve the request context (may call executor)

Time to process prepended user messages (sendUpdate + createAgentTurn per message)

Time for system prompt generation, UserInfo, tool generation, and message assembly

Time for the main stateHandler.createAgentTurn call (includes processSelectedContext)

Time for optional computeNewStructure + onStateUpdate after turn creation

 after ${e}ms

Error: Shell command was interrupted by the user${n} before it completed.

Error: Shell command was interrupted by the user${n} before it completed.

${Fs(t,{heading:"Output collected before interruption",truncatedSuffix:" (truncated)"})}

Output collected before interruption

Error: Await was interrupted by the user${TR(e)}.

Error: ${e.toolName} was interrupted by the user${TR(e.elapsedMs)} before it completed.

User message is required

dropped excess prepended user messages

prepending user messages

skipping duplicate prepended user message

${s}

${function(e,t={}){return`<system_reminder>\n${Xg(e,t)}\n</system_reminder>`}(t,r)}

<system_reminder>
${Xg(e,t)}
</system_reminder>

Failed to persist checkpoint for new turn

<system_reminder>
  IMPORTANT: Provide a status update of what you are doing.
</system_reminder>

[progress-update-reminder] injecting progress update reminder

Use `UpdateCurrentStep` to report the current major subtask to the user-facing parent timeline.

Call `UpdateCurrentStep` as your first action before doing substantive investigation or implementation work; when possible, run it in parallel with your first real work tool call.

Update `current_step` again whenever you switch to a new major subtask or phase.

Do not report tiny implementation details, routine retries, or mechanical follow-ups; keep updates high-level and user-friendly.

Keep `current_step` concise (6 words or less) and start it with a descriptive verb.

Fill `final_summary` and `completed_subtitle` once per response as your last action before the final response: a concise 1-3 sentence executive summary with the most relevant takeaways for the user, plus a 4-6 word past-tense subtitle for the UI.

Example updates: `Investigating <bug description>`, `Building <feature description>`, `Exploring <feature>`, `Testing changes`, `Refactoring <component name>`.

Use `sendFinalSummary` to capture a concise, high-signal TL;DR before your final response.

Call `sendFinalSummary` exactly once as your last tool call, right before your final response.

You may skip this tool when your final response is already very short (about 3 sentences or fewer).

Follow the tool instructions precisely.

Target vibe: high-signal, low-fluff, results-oriented, and “show, don't tell”.

Write as though you were composing a concise Slack update to your startup's CTO: lead with what matters, keep them in the loop, avoid going too low-level, and be honest if you came up short.

Keep it concise. Avoid process narration or unnecessary implementation minutiae.