[projman] Agent completion status is misleading when blocked #235

Closed
opened 2026-01-28 15:22:32 +00:00 by lmiranda · 0 comments
Owner

Problem

Agent #228 reported "completed" but actually gave up due to permission errors. The completion notification hid the actual failure.

Current Behavior

  • Agent hits permission block repeatedly
  • Eventually stops trying and reports "completed"
  • Result message contains the code it WANTED to write but couldnt
  • User thinks task succeeded when it actually failed

Expected Behavior

Clear distinction between:

  • Completed successfully - All acceptance criteria met
  • Completed with blockers - Stopped due to permissions/errors, partial work done
  • Failed - Could not make progress
  • Timeout - Hit max turns/time limit

Suggested Solutions

  1. Structured completion status - Return status enum, not just message
  2. Acceptance criteria check - Agent verifies criteria before reporting complete
  3. Error detection - Count permission denials, flag as blocked not complete
  4. Summary validation - Check if claimed work actually exists on disk

Acceptance Criteria

  • Completion status accurately reflects outcome
  • Blocked agents are clearly identified
  • User can distinguish success from failure
## Problem Agent #228 reported "completed" but actually gave up due to permission errors. The completion notification hid the actual failure. ## Current Behavior - Agent hits permission block repeatedly - Eventually stops trying and reports "completed" - Result message contains the code it WANTED to write but couldnt - User thinks task succeeded when it actually failed ## Expected Behavior Clear distinction between: - **Completed successfully** - All acceptance criteria met - **Completed with blockers** - Stopped due to permissions/errors, partial work done - **Failed** - Could not make progress - **Timeout** - Hit max turns/time limit ## Suggested Solutions 1. **Structured completion status** - Return status enum, not just message 2. **Acceptance criteria check** - Agent verifies criteria before reporting complete 3. **Error detection** - Count permission denials, flag as blocked not complete 4. **Summary validation** - Check if claimed work actually exists on disk ## Acceptance Criteria - [ ] Completion status accurately reflects outcome - [ ] Blocked agents are clearly identified - [ ] User can distinguish success from failure
lmiranda added the Component/BackendType/FeaturePriority/High labels 2026-01-28 15:23:44 +00:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: personal-projects/leo-claude-mktplace#235