On this page
Possible endings
A run should not end just because the chat stopped.
It ends by proving completion, handing off a real blocker, or refusing unsafe or unbounded work.
Audit gaps are repair loops, not final endings.
What to look for
AUDIT_COMPLETE must appear before IMPLEMENTAUDIT_RUN_COMPLETE.
If it cannot, the run should leave a handoff or stop record with the blocker and next owner action named.
Verified complete
The final audit found final, checkable evidence for every in-scope item.
The run may then emit AUDIT_COMPLETE, followed by IMPLEMENTAUDIT_RUN_COMPLETE.
AUDIT_VERIFY checks and deliverables recheckedAUDIT_COMPLETE every in-scope item has final checkable evidenceIMPLEMENTAUDIT_RUN_COMPLETE valid only after audit completionAudit gap repair loop
The final audit found a gap, but the run had a bounded fix path.
It records the gap, applies the fix, reruns the relevant evidence, and repeats the final audit loop.
AUDIT_GAPS missing or weak evidence foundcountermeasure chosen with a changed approach or new evidence
AUDIT_VERIFY rerun before closureBlocked handoff
The run cannot honestly finish without an owner decision, missing credential, unavailable environment, or scope decision.
It should name the blocker and the smallest next owner action.
AUDIT_HANDOFF not completeblocker: owner decision required
next owner action: approve route or provide missing evidence
Unsafe or unbounded stop
If the request asks for:
- unauthorized mutation
- release, publication, or provenance
- destructive filesystem behavior
- an unbounded scope jump
the run stops instead of inventing authority.
No ending on this page grants those actions by implication.
How to inspect it
.IMPLEMENTAUDIT/runs/<task-slug>-<id>/.STATE.md; for compact work, check the final audit ledger, cited artifacts, and transcript markers.End-state matrix
| Ending | What you should see | What it means |
|---|---|---|
| Verified complete | AUDIT_COMPLETE before IMPLEMENTAUDIT_RUN_COMPLETE | In-scope items have final checkable evidence. |
| Audit gap repair loop | AUDIT_GAPS, new evidence, then audit again | The final check found a gap and looped back through repair before any ending. |
| Blocked handoff | AUDIT_HANDOFF or ANDON_HANDOFF | A real blocker remains and the next owner action is named. |
| Unsafe stop | STOP or handoff text | The request is unbounded, unauthorized, or unsafe to mutate. |
Completion anti-patterns
- A stopped chat is not valid closure.
- A passing-looking rerun without cause analysis is not valid closure.
- A hidden substitution is not valid closure.
- A completion marker after a handoff is not valid closure.