QA 情况模糊时，将自然语言问题路由到最精确的 1-2 个…

用一两句话描述 QA 问题或目标，判断应调用哪些 QA 技能以及先后顺序，避免选错技能或直接硬做。

适用场景

解决思路

1. 如果调用时带有参数，把参数作为当前 QA 情况 `$ARGUMENTS`；否则先让用户用一两句话描述要做什么或遇到什么问题。
2. 根据描述匹配可识别模式，输出推荐的 1-2 个技能（按顺序），并为每个技能写一行角色说明。示例：'Our E2E tests keep failing in CI but pass locally.' → `test-reliability`（诊断 flaky/environment-sensitive tests），然后 `ci-cd-integration`（对齐 pipeline 环境与本地行为）。
3. 路由到最精确的技能，而不是最近邻。用请求中的确切 artifact 匹配技能触发短语：'Reproduce this bug' 到 `bug-reproduction`，不要到更宽的 `ai-bug-triage`；'test our Stripe checkout' 到 `payment-testing`，不要到通用 `api-testing`。不确定时优先触发短语命名了请求中确切产物的技能。
4. 最多输出两个技能，并排序——绝不堆一堆。第二个技能只有在第一个留下明确缺口时才有资格（例如 diagnose → measure，risk → checklist）。如果一个技能覆盖请求，标注 'Direct' 并停止。路由到三个技能意味着情况不明确——改为问一个澄清问题。
5. 把路由表当作事实来源，并且它会过时。每当 `skills/` 中新增或删除技能时，针对 `skills/` 重新生成路由表和 Skill Categories 参考。如果请求没有对应行，回退到 Quick Reference 中最接近的类别，而不是强行做错误的直接匹配。
6. 按 Common Situations 表执行路由（完整表）： - 新项目，完全没有测试 → `qa-start`（或 `qa-project-context` → `test-strategy`）；顺序：引导整个 QA 设置。 - 让 QA 工程师加入现有代码库 → `qa-project-bootstrap`；Direct — 30-day ramp + architecture audit。 - 'Tests keep breaking in CI' → `test-reliability` → `ci-cd-integration`；先可靠性，后 pipeline。 - 'A UI refactor/redesign broke many selectors' → `selector-drift-recovery`；Direct — bulk regen, not per-test healing。 - 'What should we test before this release?' → `risk-based-testing` → `release-readiness`；先风险，后 checklist。 - 'We need Playwright tests' → `playwright-automation`；Direct。 - 'We need Cypress tests' → `cypress-automation`；Direct。 - 'Let an agent explore the app and assert outcomes' → `agentic-browser-testing`；Direct — goal-driven, no script。 - 'Write tests from this PRD/spec/story' → `ai-test-generation`；Direct。 - 'Reproduce this vague bug / turn this report into a failing test' → `bug-reproduction`；Direct — verified minimal repro + regression test。 - 'Review my tests / find test smells' → `ai-qa-review`；Direct。 - 'Our test suite is slow and flaky' → `test-reliability` → `qa-metrics`；先诊断，后度量。 - 'Our suite is bloated / too many redundant tests' → `test-suite-curation`；Direct — audit + prune with evidence。 - 'Manage / author manual test cases (TestRail, Xray, Zephyr, Qase)' → `test-case-management`；Direct。 - 'Set up test reporting' → `qa-dashboard` → `ci-cd-integration`；先 dashboard design，后 CI wiring。 - 'Test our API' → `api-testing`；Direct。 - 'Write unit tests / add a mock / coverage threshold' → `unit-testing`；Direct。 - 'Test our database / migration / data integrity' → `database-testing`；Direct。 - 'Set up consumer-driven contract tests (Pact)' → `contract-testing`；Direct。 - 'Where are our coverage gaps?' → `coverage-analysis`；Direct。 - 'Do a structured exploratory / charter-based testing session' → `exploratory-testing`；Direct。 - 'Check accessibility compliance' → `accessibility-testing`；Direct。 - 'Test our payment / Stripe checkout / 3DS / subscription billing' → `payment-testing`；Direct。 - 'Test the signup confirmation / password reset / OTP email flow' → `email-testing`；Direct。 - 'Verify analytics / GA4 / pixel / dataLayer events fire correctly' → `analytics-tracking-testing`；Direct。 - 'We got a bug in prod, understand why' → `bug-reproduction` → `quality-postmortem`；先复现，后 retro。 - 'Classify / triage a batch of CI failures' → `ai-bug-triage`；Direct — batch failure clustering。 - 'We're migrating from Selenium/Cypress' → `test-migration`；Direct。 - 'Performance is degrading' → `performance-testing` → `observability-driven-testing`；先测量，后 trace。 - 'Set up test data' → `test-data-management`；Direct。 - 'Set up / containerize a test environment or staging' → `test-environments`；Direct。 - 'Add tests to CI' → `ci-cd-integration`；Direct。 - 'Visual changes breaking tests' → `visual-testing` → `test-reliability`；先 baseline，后 stabilize。 - 'We have no idea what quality looks like' → `qa-metrics` → `qa-dashboard`；先定义 KPIs，后呈现它们。 - 'Third-party API is unreliable in tests' → `service-virtualization`；Direct。 - 'Need to test on multiple browsers' → `cross-browser-testing`；Direct。 - 'Need to test on real iOS/Android devices (Appium/Detox/Maestro)' → `mobile-testing`；Direct。 - 'Security audit coming up' → `security-testing`；Direct。 - 'Tests depend on each other and break in random order' → `test-data-management` → `test-reliability`；先修复数据隔离。 - 'Roll out a feature safely (flags, canary) during release' → `testing-in-production`；Direct。 - 'Schedule probes / SLA checks that run after release' → `synthetic-monitoring`；Direct。 - 'Our QA is only catching bugs after dev, too late' → `shift-left-testing` → `test-planning`；先流程变更，后计划。 - 'We're building an AI/LLM feature and need to test it' → `ai-system-testing`；Direct。 - 'Make this test report sound human / less AI-y' → `qa-report-humanizer`；Direct。 - 'Make sure this is GDPR/EAA/AI Act compliant' → `compliance-testing`；Direct。 - 'Run chaos / failure injection on staging' → `chaos-engineering`；Direct。
7. 当描述同等映射到三个或更多技能时，问一个澄清问题把范围缩到 1-2 个技能。可用问题：'Are you fixing something broken, or building new coverage from scratch?'；'Is this a process problem (how the team works) or a tooling problem (what's running)?'；'Is the priority speed of delivery, or confidence in correctness?'；'Are you the only QA, or is this a team-wide change?'
8. 消解四个重叠对： - `test-reliability` vs `selector-drift-recovery`：运行时修复一个 flaky test → `test-reliability`；计划中的 UI refactor 导致大量 selectors 损坏 → `selector-drift-recovery`。 - `cross-browser-testing` vs `mobile-testing`：browsers/CSS engines → `cross-browser-testing`；真实设备、Appium/Detox/Maestro、gestures/deep links → `mobile-testing`。 - `ai-bug-triage` vs `bug-reproduction`：对一批 CI failures 分类/去重 → `ai-bug-triage`；复现并理解一个具体 bug → `bug-reproduction`。 - `qa-start` vs `qa-project-bootstrap`：全新项目、尚无 QA → `qa-start`；让 QA 工程师加入现有代码库 → `qa-project-bootstrap`。
9. 维护 Skill Categories Quick Reference：每当技能变化，从每个 `skills/*/SKILL.md` 的 `category:` 字段重新生成。当前分类为：Foundation: qa-project-context, qa-start, qa-do；Strategy: test-strategy, test-planning, risk-based-testing, exploratory-testing；Automation: playwright-automation, cypress-automation, api-testing, unit-testing, mobile-testing, visual-testing, performance-testing, cross-browser-testing, database-testing, security-testing, selector-drift-recovery；Specialized: accessibility-testing, payment-testing, email-testing, analytics-tracking-testing；AI-QA: ai-test-generation, ai-bug-triage, bug-reproduction, test-reliability, ai-qa-review, agentic-browser-testing；Infrastructure: ci-cd-integration, test-environments, test-data-management, contract-testing, service-virtualization；Metrics: qa-metrics, qa-dashboard, coverage-analysis；Process: shift-left-testing, qa-project-bootstrap, release-readiness, quality-postmortem, compliance-testing, qa-report-humanizer, test-case-management, test-suite-curation；Production: testing-in-production, synthetic-monitoring, observability-driven-testing；Knowledge: ai-system-testing, chaos-engineering, test-migration。
10. 检查并避免路由反模式：不要路由到宽泛技能而忽略精确技能；不要在一次情况上堆三个或更多技能；不要信任过时表；不要抓取属于具名兄弟技能的请求。
11. 相关技能处理：`qa-start` 是最容易与本技能混淆的兄弟技能——全新项目且尚无 QA 时用它，而不是 qa-do，它会串联 context → strategy → planning；`qa-project-context` 在多数技能使用前先捕获项目 setup，路由到它，不要重新实现；当需求是 'we need a QA strategy' 而不是一个具体问题时，用 `test-strategy`。

适用边界 · 注意事项

• 请求明确匹配另一个技能的触发短语时，直接调用该技能，不要经过 qa-do。
• 不要用于：全新项目且尚无 QA 的 bootstrapping——用 qa-start；不要用于：捕获项目 setup/context——用 qa-project-context。
• 如果描述同等映射到 3 个或更多技能，不要硬路由；先问一个澄清问题，缩小到 1-2 个技能。