ccb-self-chain - SKILL.md Agent Skill

name: ccb-self-chain description: Diagnose and repair CCB ask/job/message/reply/artifact/callback lineage. Use for missing replies, incomplete artifacts, pending callbacks, retry/resubmit/ack decisions, reply delivery problems, or work-chain resume advice.

CCB Self Chain

Use this skill for message and work-lineage recovery. Do not restart or clear a pane as the first repair for a lineage problem.

Core Workflow

Identify the lineage id: job_id, message_id, attempt_id, reply_id, inbound event id, or artifact path.
Run ccb trace <id> and record message, attempt, reply, event, callback, and job state.
Inspect observer state as supplementary evidence:
- ccb queue --detail <agent|all>
- ccb pend --inbox --detail <agent>
If a reply or request is artifact-backed, read the full artifact file before acting. If the file is absent, expired, truncated, or unreadable, report a blocker and do not act from preview text.
Classify the repair:
- retry: same job or attempt should be retried because the attempt is failed, incomplete, or retryable and the original lineage remains valid.
- resubmit: original message should be submitted as fresh work because the old lineage is no longer appropriate.
- ack: acknowledgement or inbox progress is wrong, but the reply is otherwise accepted.
If trace shows a job still running or in flight, run ccb cancel <job_id> first when the user intended repair, then re-run trace before choosing retry or resubmit. If cancel fails or reports a blocking state, stop and report the blocker; do not retry, resubmit, or create a second concurrent path for the same work.
Execute a repair command only when trace evidence supports it and the user supplied maintenance intent:
- ccb repair retry <job_id|attempt_id>
- ccb repair resubmit <message_id>
- ccb repair ack <agent_name> [inbound_event_id]
After repair, re-run trace and queue/inbox checks. Success means the new or repaired lineage is queued/running/completed as intended, the old duplicate path is cancelled or terminal, and no unexpected pending reply remains. Hand work back to the original target agent unless the user explicitly retargets it.

Decision Rules

Prefer repair retry when the same target and same work should run again after incomplete, failed, or recoverable execution.
Prefer repair resubmit when the old execution lineage is stale, context-corrupted, or semantically wrong but the user still wants the task attempted.
Prefer repair ack only for bad acknowledgement/progress state after a reply has been accepted.
For project_shutdown lineage, do not use ack. Explain that the project stopped the old execution. Prefer repair resubmit <message_id> when the user wants fresh work because repair retry is usually invalid after project shutdown. Use repair retry <job_id|attempt_id> only when trace explicitly proves the job completed before shutdown and the reply is intact.
Handoff to ccb-self-recover only when trace proves process/context/pane replacement is required.
Handoff to ccb-self-diagnose when the target id is ambiguous or runtime state is too uncertain to pick a lineage repair.

Artifact Rules

Artifact-backed replies are full-text authority for the reply content. Inline previews are hints only.

When a reply says "read the full text file above before acting":

Open the path.
Verify it is complete enough for the requested decision.
If expected checksum/bytes are available, compare them when cheap.
Only then summarize or act.

If the path is missing or expired, say that the artifact content is unavailable and choose no mutation unless the user provides the full content again.

Red Lines

Do not run ccb restart, ccb clear, or ccb reload as the first response to a lineage problem.
Do not treat queue/inbox observer snapshots as terminal authority for completed lineage; use ccb trace <id>.
Do not repair from artifact preview text alone.
Do not retarget work to another agent unless the user asks for that.