메뉴
Use when running the Rundown end-to-end test runbook and reporting structured feedback on the workflow.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Use when stepping through a Rundown runbook that is already active or has just been started, when receiving delegation instructions with a claim token, or when rundown CLI commands appear in step output. For cold-start "run the X runbook" requests with nothing active, the rundown launcher skill starts the runbook first.
Use when orchestrating multi-agent work through rundown delegation, dispatching substeps to child agents, or managing delegation tokens
Use when running the Rundown end-to-end test runbook and reporting structured feedback on the workflow.
Use when implementing a written plan task-by-task — the per-task cycle, commit discipline, and escalation rules that an execute-plan runbook orchestrates around.
Use when running the full plan → review → execute pipeline — write the plan, review it, then implement it behind review and verify gates. The top-level entrypoint that orchestrates writing-plans, plan review, and executing-plans end to end.
Use when asked to run or start a Rundown runbook by name (e.g. "run the planning runbook", "start the deploy runbook") and no runbook is active yet. The launcher that resolves a runbook and begins execution. Invocable as /rundown <runbook>.
| name | end-to-end-testing |
| description | Use when running the Rundown end-to-end test runbook and reporting structured feedback on the workflow. |
Run the end-to-end test through Rundown state. Do not improvise. Start:
rundown run rundown:end-to-end-test
end-to-end-test/write-file.runbook.md, and
end-to-end-test/review-and-collate.runbook.md run locally.end-to-end-test/review-file.runbook.md and
end-to-end-test/collate-files.runbook.md are delegated by the review
wrapper.Only nested review and collation runbooks are delegated. If the wrapper completes before collate, report a defect; never delegate collation from the parent.
--text (it is
human-only output, not part of the agent protocol).rundown run, rundown pass, rundown fail, rundown claim, and
rundown collect output as the next context. Parent-side pass/fail/collect
carry --run <rd_…> (capture the run id from rundown run output, echoed as
runbookId on every event); child-side commands carry
--claim-id <claim_id>.rundown status only to recover orientation after an error or
interruption..rundown.The DELEGATE step auto-issues a claim token (delegations in rundown status).
Claim it:
rundown claim <token> # returns claim_id; dispatch the child to a subagent
Drive the claimed child like any runbook: advance each step with
rundown pass --claim-id <claim_id> / rundown fail --claim-id <claim_id>.
Prompted child steps need this claim-id transition to advance. On a
delegation-exposed run, a bare rundown pass/rundown fail is refused with
ACTOR_CONTEXT_REQUIRED regardless of claim state (exposure is sticky). A
--run-targeted parent advance is additionally refused with
OPEN_DELEGATED_CHILDREN while a claimed child is open.
When the child's final step completes it auto-resolves its parent substep and
the parent auto-aggregates and advances. rundown pass/fail --claim-id is
idempotent on a resolved child, so it also confirms or overrides a child you
stop early.
Use rendered {{ path Alias }} paths.
Expected artifacts:
PlanSchemaPathPlanPathReviewSchemaPathReviewPathReviewPathsCollatedReviewPathFeedbackPathDo not rediscover paths or add artifact-only steps. Missing/wrong variables and write-one/check-another mismatches are defects.
Feedback must be JSON matching review.schema.json. Use empty items when
there are no findings.
Capture only concrete issues: unclear prompts, wrong delegation boundary, missing/wrong artifact variables, schema mismatches, required manual inference, failed commands with observed output.