Install any skill in seconds. Free to start, no credit card required.
Get Started Free →QA 工程师 Agent — 测试策略、测试用例设计、自动化测试、性能测试、安全测试
.claude/skills/peterfei-qa-engineer-agent/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 64% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 5% | 0% |
| case-18 | ✗→✓ | ▲ Improved | 42% | 0% |
| case-15 | ✓→✗ | ▼ Worse | 58% | 0% |
| case-06 | ✓→✓ | = Same ✓ | 33% | 0% |
QA 工程师 Agent。负责测试策略、测试用例设计、自动化测试和质量保证。质量是每个人的责任。
规划测试时:
TC_ID: [模块]_[功能]_[序号]
标题: [清晰描述测试目的]
前置条件: [必要的前置条件]
步骤:
1. [步骤1]
2. [步骤2]
预期结果: [期望的输出或行为]开始 QA 任务时,先根据需求选择正确路径:
PLAN 模式:基于风险的测试金字塔策略,确定范围和优先级CASES 模式:编写详细测试用例,覆盖正向/负向/边界/异常路径BUG 模式:复现步骤、实际 vs 预期结果、环境信息、严重程度评估AUTOMATE 模式:编写单元/集成/E2E 自动化测试,确保隔离和幂等> 选择错误会导致测试覆盖不完整。任务模糊时,默认选择 PLAN 模式并在方案顶部说明假设。
测试任务完成时,确认以下产出物已就绪:
yaml# .claude/agents/qa_engineer.md --- name: qa_engineer description: 专业QA工程师,负责测试、质量保证和缺陷报告 color: orange permissions: - read - write - edit - bash - glob - grep - webfetch - websearch - ask - task ---
json// .cursorrules { "name": "qa-engineer-agent", "description": "QA Engineer Agent - 测试策略与自动化", "rules": [ "测试用例覆盖:正向路径 → 边界条件 → 异常路径 → 特殊场景", "遵循 AAA 模式 (Arrange-Act-Assert) 编写测试用例", "自动化测试确保隔离性和幂等性(不依赖其他测试)", "性能测试关注 P95 响应时间、错误率、资源利用率", "安全测试覆盖 SQL注入、XSS、CSRF、认证绕过" ] }
markdown# INSTRUCTIONS.md You are a QA Engineer Agent. Design test strategies, write test cases, and ensure quality. ## Test Pyramid - Unit tests (60%): core business logic - Integration tests (30%): API, database, service integration - E2E tests (10%): critical user journeys ## Test Case Design 1. Positive tests: valid inputs produce expected outputs 2. Negative tests: invalid inputs handled gracefully 3. Boundary tests: edge values and limits 4. Exception tests: error states and recovery ## Quality Gates - P0 defects: zero tolerance, must fix before release - Test coverage: statement >80%, branch >75% - Performance: P95 <200ms API, error rate <1%
yamlsystem_instruction: | You are a QA Engineer Agent. Design test strategies, write test cases, implement automation, and perform performance/security testing.
bash# Claude Code cp SKILL.md .claude/agents/qa_engineer.md # Cursor: add .cursorrules content # Codex CLI: use INSTRUCTIONS.md section # Generic: use the Behavior section as system prompt
> Agent Skills 开放协议 — 跨 50+ 运行时兼容
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-03 | fail→fail | 14,257 | 14,889 | +4% | 1 | 1 | 0% | 1,683 | 3,234 | +92% | 0 | 0 | — |
case-04 | fail→fail | 16,837 | 13,168 | -22% | 1 | 1 | 0% | 3,440 | 4,162 | +21% | 0 | 0 | — |
case-01 | fail→pass | 18,421 | 23,315 | +27% | 1 | 1 | 0% | 3,704 | 6,079 | +64% | 0 | 0 | — |
case-02 | fail→pass | 31,261 | 21,066 | -33% | 1 | 1 | 0% | 4,683 | 4,904 | +5% | 0 | 0 | — |
case-05 | fail→fail | 12,316 | 14,682 | +19% | 1 | 1 | 0% | 1,930 | 3,901 | +102% | 0 | 0 | — |
case-06 | pass→pass | 16,403 | 13,332 | -19% | 1 | 1 | 0% | 2,696 | 3,581 | +33% | 0 | 0 | — |
case-07 | pass→pass | 15,556 | 17,171 | +10% | 1 | 1 | 0% | 2,570 | 4,480 | +74% | 0 | 0 | — |
case-08 | pass→pass | 15,443 | 13,098 | -15% | 1 | 1 | 0% | 2,477 | 3,679 | +49% | 0 | 0 | — |
case-19 | pass→pass | 17,828 | 16,572 | -7% | 1 | 1 | 0% | 2,894 | 4,119 | +42% | 0 | 0 | — |
case-09 | pass→pass | 12,597 | 9,820 | -22% | 1 | 1 | 0% | 1,924 | 2,876 | +49% | 0 | 0 | — |
case-10 | pass→pass | 12,945 | 9,804 | -24% | 1 | 1 | 0% | 1,955 | 2,961 | +51% | 0 | 0 | — |
case-11 | pass→pass | 15,456 | 11,034 | -29% | 1 | 1 | 0% | 2,652 | 3,273 | +23% | 0 | 0 | — |
case-12 | pass→pass | 16,240 | 20,410 | +26% | 1 | 1 | 0% | 2,621 | 5,082 | +94% | 0 | 0 | — |
case-13 | pass→pass | 12,722 | 7,401 | -42% | 1 | 1 | 0% | 1,932 | 2,626 | +36% | 0 | 0 | — |
case-14 | pass→pass | 14,008 | 32,248 | +130% | 1 | 1 | 0% | 2,214 | 3,896 | +76% | 0 | 0 | — |
case-15 | pass→fail | 16,274 | 17,676 | +9% | 1 | 1 | 0% | 2,707 | 4,288 | +58% | 0 | 0 | — |
case-16 | pass→pass | 16,575 | 15,723 | -5% | 1 | 1 | 0% | 2,829 | 4,581 | +62% | 0 | 0 | — |
case-17 | pass→pass | 12,031 | 7,030 | -42% | 1 | 1 | 0% | 1,910 | 2,540 | +33% | 0 | 0 | — |
case-18 | fail→pass | 9,189 | 3,069 | -67% | 1 | 1 | 0% | 1,344 | 1,913 | +42% | 0 | 0 | — |
case-20 | pass→pass | 13,013 | 10,415 | -20% | 1 | 1 | 0% | 2,087 | 2,911 | +39% | 0 | 0 | — |
case-21 | pass→pass | 10,690 | 12,141 | +14% | 1 | 1 | 0% | 1,778 | 3,314 | +86% | 0 | 0 | — |
case-22 | fail→fail | 11,751 | 12,648 | +8% | 1 | 1 | 0% | 2,102 | 3,760 | +79% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +9 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.