| name | autopilot-test-health |
| description | Use when the user wants to quantify test effectiveness — running lightweight shell-layer mutation testing and static health scoring to confirm the test suite would actually catch regressions. Typical triggers include weekly CI sweeps, on-demand audits asking 'how healthy are our tests', or '/autopilot-test-health score|mutate|all' invocations. Not part of any phase gate. |
| user-invocable | true |
| argument-hint | [score|mutate|all] — 默认 all |
Autopilot Test Health — 测试有效性量化
用途：在已有 autopilot-test-audit（静态扫描候选清单）之上，再加一层"用变异测试量化测试有效性"的能力。避免测试存在但无信号。
何时使用
- CI weekly sweep：定时跑
/autopilot-test-health all 输出 dashboard
- 人工 audit：发版前评估测试套件健康度
- 新手定位弱测试：top-10 weak files 直接给出改进目标
不建议挂 pre-commit：变异测试每次至少跑 N 个 mutant × test，开销太大。
模式
| 模式 | 含义 | 输出 |
|---|
mutate | 仅跑变异测试 | .cache/spec-autopilot/mutation-report.json |
score | 仅跑静态健康度评分（含读入已有 mutation 结果） | .cache/spec-autopilot/test-health-report.json |
all（默认） | 先 mutate 再 score | 两份报告 |
执行步骤
Step 1: 解析参数
MODE="${1:-all}"
case "$MODE" in
mutate|score|all) ;;
*) echo "用法: /autopilot-test-health [score|mutate|all]"; exit 1 ;;
esac
Step 2: 执行变异测试（mutate / all）
bash ${CLAUDE_PLUGIN_ROOT}/runtime/scripts/test-mutation-sample.sh \
--targets "plugins/spec-autopilot/runtime/scripts/*.sh" \
--sample-size 10 \
--timeout-per-mutant 30
输出：
.cache/spec-autopilot/mutation-report.json（含 targets / overall_kill_rate / survivors）- stdout:
MUTATION_KILL_RATE=0.XX SURVIVORS=N
Step 3: 执行静态评分（score / all）
bash ${CLAUDE_PLUGIN_ROOT}/runtime/scripts/test-health-score.sh \
--tests-dir plugins/spec-autopilot/tests \
--threshold 60
输出：
.cache/spec-autopilot/test-health-report.json- stdout:
HEALTH_SCORE=X 与可选 HEALTH_BELOW_THRESHOLD=1
Step 4: 总结
读取两份报告，汇总 dashboard：
Mutation kill_rate: 0.XX (survivors: N)
Health overall: XX/100
assertion_density: 0.XX
weak_ratio: 0.XX
duplicate_ratio: 0.XX
kill_rate: 0.XX (来自 mutation 报告)
安全约束
- 要求 git working tree 干净：变异测试会临时编辑源文件，脏 tree 直接 exit 2
- 变异 + 恢复后 git diff 必须为空：双重校验
- 超时强制 kill：避免 sleep / 死循环阻塞
- 失败不阻断 CI：两个脚本都 exit 0；阈值命中仅以
HEALTH_BELOW_THRESHOLD=1 提示
输出消费建议
kill_rate < 0.5 → 该批 runtime 脚本测试质量整体不达标weak_ratio > 0.3 → 大量测试只断言 exit code，需补内容/格式断言duplicate_ratio > 0.2 → case 命名重复，可能是复制粘贴未改名top_weak 列表中的文件 → 优先改进对象
配置
.claude/autopilot.config.yaml 可加：
test_health:
thresholds:
overall: 60
assertion_density: 2.0
weak_ratio_max: 0.3
参考
- 变异规则：
references/mutation-strategies.md
- 健康指标：
references/health-metrics.md
