AgentSkillsCN

ap-check

依据既定的检查项对已完成的工作进行验证,记录相关证据,判定通过或未通过,并在检查不达标时及时添加修复任务。可在执行 /ap:do 之后,或在确认工作完成前使用。

SKILL.md
--- frontmatter
name: ap-check
description: Verify completed work against defined checks, record evidence, mark pass or fail, and add fix tasks when checks fail. Use after /ap:do or before calling work done.

/ap:check

Purpose

Verify that completed work actually meets expectations.

/ap:check runs acceptance checks, records results, and determines what happens next. It is the only command that can mark checks as pass or fail.


Inputs

Any of the following:

  • One or more check IDs (e.g. C001 C002)
  • One or more task IDs (checks are inferred)
  • No input (defaults to all active checks)

Optional:

  • Notes on how checks should be run (e.g. "manual review acceptable").

Reads

Must consult:

  • .agent-pack/work/CHECKS.md
  • .agent-pack/work/PLAN.md
  • .agent-pack/work/STATUS.md
  • .agent-pack/context/PROJECT.md

May consult:

  • Recent runs/ output
  • Repo source code

Process

  1. Select checks
    • Resolve input into a concrete list of checks.
    • Only checks from the active plan may be run.
  2. Re-read intent
    • Review the task's done-when condition and plan goal.
  3. Run the check
    • Follow the "How to verify" steps.
    • Record evidence.
  4. Record result
    • Mark each check as pass or fail.
    • Move failed checks to the Failed section.
  5. Handle failures
    • For each failed check:
    • Identify failure mode
    • Propose one or more fix tasks
    • Each fix task must be small and map to a new check (or reuse the failed one)
    • Add fix tasks directly to PLAN.md under a clearly marked "Fix tasks" section.
    • Do not implement fixes during verification unless explicitly instructed.
  6. Update status
    • Update STATUS.md to reflect progress.
    • If all checks pass, mark the plan complete.

Output rules

  • Follow OUTPUT CONTRACTS.
  • Checks must not be redefined during verification.
  • Evidence should be sufficient for later review.

Output contract

md
### Files updated

- .agent-pack/work/CHECKS.md -- <summary>
- .agent-pack/work/STATUS.md -- <summary>
- .agent-pack/work/PLAN.md -- <summary if status changed>

### Decisions made

- (rare; only if verification forces a durable change)

### Tasks created or modified

- <fix tasks added to PLAN.md, if any>

- <new fix tasks, if any>

### Checks created or affected

- C### -- pass / fail

### Unknowns / needs input

- <questions or blockers>

Completion criteria

/ap:check is complete when:

  • Each selected check is marked pass or fail
  • Evidence is recorded for each result
  • STATUS.md reflects the current state

If failures exist, next actions must be explicit.


Common failure modes

  • Marking checks as pass without evidence
  • Quietly changing checks to make them pass
  • Fixing issues during verification without visibility
  • Leaving ambiguous or partial results

Notes

Verification is not a formality.

If checks frequently fail, improve planning or task scoping -- not the checking process.