| name | task-audit-adjudication |
| description | Use when main_brain needs to inspect one task in depth, reconstruct its event and queue history, or compare multiple worker artifacts through adjudicate_task.sh before any human-gated close, reopen, or cancel decision. |
Task Audit Adjudication
Use this skill for structured audit, not free-chat arbitration.
Inspection chain
bash scripts/show_task.sh --task <task_id> --target <dir>
bash scripts/list_history.sh --task <task_id> --target <dir>
bash scripts/adjudicate_task.sh --task <task_id> --target <dir>
bash scripts/artifact_index.sh --target <dir>
Core rules
show_task.sh is the current-state view.list_history.sh is the timeline and audit trace view.adjudicate_task.sh is the structured comparison draft.artifact_index.sh is a derived review-side index when artifacts are scattered.- Final state change still belongs to main brain through repo scripts.
Evidence handling
- Default adjudication compares worker-produced evidence first.
- Callback or event-derived artifacts are supporting evidence, not primary truth.
- If evidence is incomplete, say so explicitly instead of forcing a verdict.
Do not do
- Do not auto-close a task from adjudication output.
- Do not use free chat as the primary comparison mechanism.
- Do not treat
.codex as a second state machine.
