verify-a-child-agents-output
Use when a subagent or helper reports its task complete. Re-runs the acceptance check and inspects the artifact against the brief, because a child's "done" is a claim to audit, not evidence.
- License 许可
- MIT
- 最近更新
- 2026-11-13
- Stars
- ★ 81
- Forks
- 17
01 前言 Front Matter
- name
- verify-a-child-agents-output
- description
- Use when a subagent or helper reports its task complete. Re-runs the acceptance check and inspects the artifact against the brief, because a child's "done" is a claim to audit, not evidence.
02 资源文件 3
- 许可证LICENSE1.0 KB
- 参考文档SOURCE.md354 B
- 参考文档TRUST.auto.yaml2.1 KB
03 SKILL.md 正文
Verify a Child Agent's Output
A child's success message is testimony, and the parent owns the result. Re-derive it: run the check the brief specified, open the artifact, and confirm the count matches the ask.
Procedure
- Re-read the brief at
notes/children/<child-id>.brief.mdso you check against what was asked, not against what the child produced. - Run the same acceptance command you handed the child, yourself:
python3 tools/validate.py --only foo. Never trust a pasted summary. - Confirm declared totals against actual ones: if the child says "20 skills", run
ls skills/ | wc -land compare rather than eyeballing. - Open the artifact at its path and read it; a file can pass a linter and still hold the wrong content.
- Check the boundary the brief set with a read-only
git status --shortand confirm no forbidden file changed. - Assert on counts with code, not prose:
grep -c '^## ' skills/foo/SKILL.mdreturns 3, not "the headings look fine". - Distinguish "ran without error" from "produced the right output" — exit 0 with empty output is a silent failure.
- If any check disagrees with the child's report, treat the whole report as unverified and re-examine each claim.
- Record the verdict as a line in
notes/verification.log:child-id PASS|FAIL command exit-code. - Re-dispatch only the failed slice, never the whole task.
Pitfalls
- Accepting a child's summary line ("20/20 valid") without re-running the command that produces it; a header can be typed by hand.
- Reading the diff the child chose to show rather than the artifact on disk, which may carry extra edits.
- Letting the child's confident tone substitute for a passing check.
- Comparing a declared count to a filtered file listing and calling it verified.
- Marking a child verified because no error printed, when the command exited 0 without emitting its expected line.
Verification
python3 tools/validate.py --only foo; echo "exit=$?"
# passes when the parent-run check reports the same result the child claimed
Report to the user: the child id, the acceptance command you re-ran, its exit code, and any claim that did not hold.
04 同仓库更多 Skills
recipe-performance
Diagnose a specific MetaMask user flow's real-app performance and Sentry critical-user-flow telemetry with deterministic mm-harness recipes. Use for CUF boundaries, cold/warm/background-resume behavior, latency regressions, lifecycle tags, duplicate or blocking network requests, long tasks, layout/paint cost, screen redraws, React rerenders, or browser trace analysis where tests alone cannot prove the real app.
mexc-spot-websocket
MEXC Spot WebSocket workflows for public streams, protobuf handling, depth maintenance, listen keys, and private streams without exposing API keys.
mexc-spot-rest
MEXC Spot REST workflows for public data, signed account requests, orders, wallet/admin actions, and secure signing without exposing API keys.
mexc-futures-websocket
MEXC Futures WebSocket workflows for public streams, private account/order/position streams, ping/pong, subscriptions, and depth maintenance.
0xklen-version-and-refresh-sanctions-lists
暂无简介
verify-tests-cover-changed-lines
Use when a PR adds code that the existing suite may not exercise. Measures patch coverage — the share of newly changed lines hit by tests — before approving.