Skip to content

fix(observer): stop re-nudging workspaces that answered with a question - #39

Merged
elitecoder merged 1 commit into
mainfrom
worktree-fix-observer-b5-misfire
Oct 7, 2026
Merged

elitecoder merged 1 commit into
mainfrom
worktree-fix-observer-b5-misfire

Conversation

@elitecoder

Copy link
Copy Markdown
Owner

4-line clarification to the Observer prompt. See branch for diff.

Add two clarifications to stop the repeated-nudge loop on needs_user workspaces:

1. B1 re-fires after a stall nudge: a workspace that responds to a nudge
   with a question or "your call" close is in B1 state. Do not re-classify
   as stranded on the next idle cycle.

2. B5's mid-narrative condition now has an explicit bullet: a turn ending
   with a question or handoff is not mid-narrative — B5 does not fire for it.

3. Cheat-sheet: add a row for the previously uncovered case (idle > 1800s +
   last assistant text is a question → needs_user, not stranded).

Root cause: workspace 17 received four stall nudges today. Each time it
responded "nothing stalled, your call: alert-only or pause?" and the Observer
correctly emitted needs_user. But on the next idle cycle (30+ min) the LLM
re-classified the same question text as mid-narrative and fired stranded again.
The NO_INGEST_GUARD stopped a fifth nudge (delta=0), but four was already too many.

Co-Authored-By: Claude Sonnet 4.6 (1M context) <[email protected]>
@elitecoder
elitecoder merged commit cfe75fc into main Oct 7, 2026
1 check passed
@elitecoder
elitecoder deleted the worktree-fix-observer-b5-misfire branch October 7, 2026 19:50
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.

1 participant