agentsclimarketplace

Verification before completion

Skill BlueprintOS/analysis-to-delivery/skills/orchestration/development/verification-before-completion

完成任务前验证(铁律)— 来自 superpowers 体系。本 skill 是桥接层,完整纪律见 <SUPERPOWERS_SKILL_ROOT>/verification-before-completion/。From its SKILL.md

Install
npx -y skills add BlueprintOS/analysis-to-delivery --skill verification-before-completion

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 24 stars24 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
  • runs commandsInstructs the agent to run 1 command, including `; echo "exit=$?"`.

SKILL.md

2.9 KB, 957 tokens by cl100k_base, as published. Nobody here has run it

Verification-Before-Completion(桥接到 superpowers)

Contract

  • Inputs: completion claim, verification commands, implementation/documentation outputs
  • Outputs: verification report using IDENTIFY/RUN/READ/VERIFY/THEN
  • Gates: full verification command succeeds and output supports the claim
  • Required disciplines: stage-gate
  • Next: commit, fix loop, or /handoff

本仓库不维护此 skill 的内容。完整纪律请读: <SUPERPOWERS_SKILL_ROOT>/verification-before-completion/SKILL.md

何时调

  • 每个子任务结束时(在 commit 之前)
  • 整个开发冲刺结束时(在 HANDOVER 之前)

衔接点

  • 产出:验证报告(IDENTIFY → RUN → READ → VERIFY → THEN 5 步门控)
  • 下一步:commit(通过)或修复(失败)
  • 门控:disciplines/stage-gate 第 3 层(失败必须重新跑)

5 步门控

参考完整纪律:

  1. IDENTIFY — 什么命令能证明这个声明?
  2. RUN — 执行完整的验证命令
  3. READ — 读完整输出,检查退出码
  4. VERIFY — 输出是否确认了声明?
  5. THEN — 才能做出声明

铁律:跳过任何一步 = 撒谎,不是验证。

降级方案(superpowers 未装时)

如果 <SUPERPOWERS_SKILL_ROOT>/verification-before-completion/ 不存在,严格按以下 5 步门控(铁律,不允许跳):

1. IDENTIFY — 什么命令能证明这个声明?

  • 把"已完成 / 通过 / 成功"翻译成可执行的命令
  • 例如:"已修复 ASN 收货 bug" → mvn test -Dtest=AsnReceiveServiceTest
  • ❌ 没有命令 = 你不知道自己在声称什么

2. RUN — 执行完整命令

  • 命令必须完整执行(不能只看 stdout 头几行就停)
  • 包含退出码 / 错误输出
  • 用 ; echo "exit=$?" 拿退出码

3. READ — 读完整输出

  • 读全部输出,不只是你期望的那部分
  • 数失败数 / 数 PASS 数 / 检查退出码
  • ❌ 看一半就跳 = 漏掉隐藏失败

4. VERIFY — 输出是否真的支持声明?

  • 把命令输出对照你的声明
  • 输出支持 → 进 THEN
  • 输出反对你的声明 → 立刻如实汇报(不允许"应该可以了")
  • ❌ "看起来是对的" 一律不算

5. THEN — 才能做出声明

  • 经过前 4 步后才能写"完成 / 通过 / 成功"
  • 必须附证据:命令 + 退出码 + 关键输出片段
  • ❌ 没跑就声明 = 撒谎

最小纪律摘要

  • IDENTIFY 写不出命令 = 这个声明不成立
  • RUN 不完整 = 假装在验
  • READ 跳着看 = 漏失败
  • VERIFY 不严 = 给自己找借口
  • THEN 没附证据 = 不可信

安装提示

npx skills@latest add obra/superpowers-verification-before-completion

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.