Wire QA intent detection in daemon routing + fix epoch units#6941
Conversation
Co-Authored-By: Claude <noreply@anthropic.com>
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: ddf847cf30
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| screenWidth, | ||
| screenHeight, | ||
| interactionType: 'computer_use', | ||
| reportToSessionId: sourceSessionId, |
There was a problem hiding this comment.
Setting reportToSessionId on every escalation stores CU metadata for every escalated run, but the macOS client only sends cu_session_finalized when qaMode is true (ComputerUseSession.run), so non-QA escalations never trigger the metadata cleanup path in handleCuSessionFinalized. In long-lived desktop sessions this accumulates stale cuSessionMetadata entries and leaves the new report-linking path effectively incomplete unless QA mode is also enabled.
Useful? React with 👍 / 👎.
| const lower = taskText.toLowerCase().trim(); | ||
|
|
||
| // Direct QA/test commands | ||
| if (/^(qa|test|verify|check)\b/.test(lower)) return true; |
There was a problem hiding this comment.
The direct matcher marks any task beginning with check as QA intent, so routine computer-use requests like “check my inbox/calendar/slack” are treated as QA/test workflows even when no testing is intended. That broad trigger can unexpectedly flip sessions into QA-specific behavior and pollute QA/reporting paths for normal user tasks.
Useful? React with 👍 / 👎.
Co-authored-by: Vellum Assistant <assistant@vellum.ai> Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Vellum Assistant <assistant@vellum.ai> Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Vellum Assistant <assistant@vellum.ai> Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Vellum Assistant <assistant@vellum.ai> Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Vellum Assistant <assistant@vellum.ai> Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Vellum Assistant <assistant@vellum.ai> Co-authored-by: Claude <noreply@anthropic.com>
1 participant
P0: Add detectQaIntent() for keyword/pattern matching of QA tasks. Wire it in task_submit and escalation handlers to set qaMode+reportToSessionId on CU sessions. Add qaMode/reportToSessionId to TaskRouted IPC message so clients know to start recording. P1: Fix epoch seconds→ms mismatch in Session.swift finalizeQARecording. P2: Map unknown session status to failed to match TS contract union.