返回全部 Skills
05 / 验收

Prove Behavior

prove-behavior

Verify that an implemented change works through observable runtime behavior and representative user paths. Use for acceptance testing, QA, release confidence, or checking a finished implementation; not for diagnosing an unexplained failure.

仓库
linonward/skills
源文件
skills/prove-behavior/SKILL.md
许可
MIT

// Install one skill

安装这个 Skill

先阅读下面的源文件;确认它适合当前工作方式后,再运行安装命令。

npx skills add linonward/skills --skill prove-behavior

// Read-only source preview

阅读源文件

Prove Behavior

Verification answers whether the promised outcome works in the environment that matters. A clean build is supporting evidence, not proof of a user journey.

Outcome Contract

Map each material acceptance claim to direct evidence, exercise the primary path and meaningful failure paths, and report what remains unverified.

Evidence Order

Prefer the closest feasible observation to the real outcome:

  1. representative UI, API, CLI, or produced artifact;
  2. integration or end-to-end behavior;
  3. focused automated tests;
  4. static checks and build success.

Lower layers may support higher layers but should not be presented as equivalent evidence.

Hard Rules

  • Derive checks from the requested behavior and changed risk, not from whichever test command is easiest to run.
  • Verify negative and boundary behavior when failure would harm data, permissions, money, or user trust.
  • Never claim an unobserved result. State environmental or credential limits explicitly.
  • Avoid changing product code merely to make verification convenient unless the user asked for implementation changes.
  • Capture enough command output, response data, screenshot, or artifact detail for another person to understand the conclusion.

Output

Return a concise evidence table with Claim, Check, Result, and Evidence, followed by uncovered risks. Lead with pass, partial pass, or fail.