[P1] Stabilize transient dirty-checkout verification #893

New Issue

2026-07-24T21:37:06-05:00

jcwalker3 commented

2026-07-24 21:37:06 -05:00

Parent epic: #887
Priority: P1

Problem

One temporary M gitea_mcp_server.py observation permanently blocked a run even though the immediate diff was empty when re-read. The check samples once and latches. Related known defect: runtime_context scans unstaged changes only, so a staged-dirty checkout reads clean — the two failure directions must be fixed together or the fix trades one for the other.

Acceptance criteria

A transient dirty observation triggers bounded revalidation rather than an immediate permanent block.
Revalidation confirms HEAD, index state, worktree diff, and file metadata — not just the unstaged diff.
The run proceeds only after a stable clean sample window.
Persistent or reproducible modifications block immediately, with no revalidation grace.
Evidence supporting either outcome is recorded and attached to the run.
Protection for genuinely dirty control checkouts is not weakened; a regression proves a real dirty checkout still blocks.

Duplicate verdict

NOT a duplicate. #697 covers preflight ignoring untracked files (the opposite direction — under-detection). #867 covers detecting out-of-band mutations. Neither addresses a single transient observation latching a run into a permanent blocked state.

Parent epic: #887 Priority: P1 ## Problem One temporary `M gitea_mcp_server.py` observation permanently blocked a run even though the immediate diff was empty when re-read. The check samples once and latches. Related known defect: `runtime_context` scans unstaged changes only, so a staged-dirty checkout reads clean — the two failure directions must be fixed together or the fix trades one for the other. ## Acceptance criteria - A transient dirty observation triggers bounded revalidation rather than an immediate permanent block. - Revalidation confirms HEAD, index state, worktree diff, and file metadata — not just the unstaged diff. - The run proceeds only after a stable clean sample window. - Persistent or reproducible modifications block immediately, with no revalidation grace. - Evidence supporting either outcome is recorded and attached to the run. - Protection for genuinely dirty control checkouts is not weakened; a regression proves a real dirty checkout still blocks. ## Duplicate verdict NOT a duplicate. #697 covers preflight *ignoring untracked files* (the opposite direction — under-detection). #867 covers detecting out-of-band mutations. Neither addresses a single transient observation latching a run into a permanent blocked state.

jcwalker3 added the status:ready workflow-hardening type:bug contamination labels 2026-07-24 21:37:07 -05:00

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: Scaled-Tech-Consulting/Gitea-Tools#893