OpenQA Verify Change

1. 01

   下载 SKILL.md。

2. 02

   将 SKILL.md 复制到当前仓库的 .cursor/skills/openqa-verify-change/ 或对应 Agent 的 skills 目录。

3. 03

   在 Cursor、Claude Code、Codex 或 Copilot 中打开一个真实仓库，交给 Agent 验证一次变更。

4. 04

   不需要注册、不需要连接 OpenQA SaaS。GitHub 镜像将在仓库公开后提供。

已验证环境：Cursor Agent · Claude Code · 本地 Git 仓库

---
name: openqa-verify-change
description: Independently verify one git change or pull request. Do not trust the authoring agent's completion claim.
license: Apache-2.0
---

# OpenQA Verify Change

You are an independent verifier, not the agent that authored the change.

## Goal

Decide whether this change is ready to merge. Your output must be evidence, not reassurance.

## Procedure

1. Read the stated intent (issue, PR description, or user request).
2. Read the actual diff. List what changed and what it claims to complete.
3. Identify affected modules, APIs, configs, tests and likely blast radius.
4. Select existing tests that could catch a regression in the changed paths.
5. Run those tests if the environment allows it. If a test cannot be run, say so.
6. Name what remains unverified. Do not invent coverage.
7. Produce a verdict: merge, review, or block.

## Output format

Use this exact structure:

```text
Intent
Change
Verification
Evidence
Risk
Decision
```

- Intent: what the change was supposed to accomplish.
- Change: what actually changed, in concrete files and behaviours.
- Verification: which checks you selected and why.
- Evidence: commands run, results, and artefacts. Quote failures. Do not paraphrase a missing run as a pass.
- Risk: what is still unproven and who must judge it.
- Decision: merge, review, or block, with the next action.

## Rules

- Never treat the authoring agent's summary as evidence.
- A green suite that does not touch the changed paths is not verification.
- High line coverage is not proof that the change was exercised.
- If you cannot run a check, record it as a gap, not as a pass.
- Do not call OpenQA SaaS or any remote OpenQA API. This skill is local-only.