Skip to content

[FIX] Work log: record failed runs + human-readable action labels - #73

Open
ScottieR wants to merge 2 commits into
masterfrom
cuj/work-log-honesty
Open

ScottieR wants to merge 2 commits into
masterfrom
cuj/work-log-honesty

Conversation

@ScottieR

Copy link
Copy Markdown
Owner

Fixes #64 — from the 2026-08-24 Sloane health-forms CUJ test: the run's fatal failure left zero trace in the Activity work log (so the log couldn't confirm or deny the agent's false "I'm on it" claim), and existing entries read as infrastructure codes.

Changes

  • Failed runs leave a trace: finalize_thread_run writes a run_failed audit entry ("Attempted a task and failed: ") — the one chokepoint all failure paths pass through. Success already logged chatted; this is its counterpart.
  • Human labels: the work log renders "Files synced" / "Task failed" / "Chat" etc., machine code preserved in the hover title. run_failed automatically picks up the red failure styling.

Testing

  • cargo check clean, tsc --noEmit clean, frontend suite passes

🤖 Generated with Claude Code

From the 2026-08-24 Sloane CUJ test (issue #64): the fatal send failure
left no trace in the Activity work log, and what the log did show read
as infrastructure ('FILES_BRIDGE_SYNCED VIA FILES — 0 folder grant(s)
delivered via brokered_read_only').

- finalize_thread_run now writes a 'run_failed' audit entry ('Attempted
  a task and failed: <error>') at the single chokepoint every failure
  path goes through — success already logged 'chatted'; failures were
  invisible, which is exactly what let a false 'I'm on it' claim go
  unchallenged by the log meant to verify it.
- ActivityTab renders human action labels ('Files synced', 'Task
  failed', 'Chat') with the machine code preserved in the hover title.
  run_failed entries pick up the existing red failed styling.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

@github-actions github-actions Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Automated review (Claude)

**Verdict: ** | **Risk: **

Blocking issues:

improvement suggestion(s) filed as issues for later triage (see repo Issues, label claude-suggestion).

…vals

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[UX] Human-readable work log that records failed attempts

1 participant