菜单
Select, execute, and report risk-based validation for a change or release candidate. Use for unit, integration, end-to-end, contract, regression, invariant, static, performance, security, CI-parity, and failure-triage evidence.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use first to route end-to-end, ambiguous, multi-phase, review, release, incident, or decommission work through the minimum sufficient project-delivery lifecycle when depth, state, or authority is unclear.
Close delivery by capturing outcomes, learning, technical debt, follow-up work, reusable decisions, and measurable process improvements in canonical project knowledge.
Coordinate delivery across authorized trackers, documents, meetings, calendars, messages, and operational systems while preserving canonical sources, exact targets, write authority, traceability, and read-after-write evidence.
Update canonical project knowledge for users, developers, operators, reviewers, releases, and future agents. Use for docs-only work or whenever a change affects behavior, interfaces, configuration, architecture, operations, migration, status, or release communication.
Prepare, validate, and—when explicitly authorized—execute integration, CI/CD, versioning, deployment, rollback, monitoring, post-release verification, and hotfix/change-management work.
Assess and improve security, privacy, risk, reliability, observability, supportability, and operational readiness across requirements, design, implementation, review, or release.
| name | testing-quality |
| description | Select, execute, and report risk-based validation for a change or release candidate. Use for unit, integration, end-to-end, contract, regression, invariant, static, performance, security, CI-parity, and failure-triage evidence. |
| license | MIT |
Read ../.shared/operating-model.md; use the test evidence template in ../.shared/artifact-templates.md.
Use after implementation, during planning/design to define strategy, when CI fails, for release gates, or for test-only/quality requests. Lead flaky-CI triage, performance-only claims, and combined platform performance/packaging/signing/distribution evidence recovery. Select depth from scale, blast radius, failure modes, and repository norms—not a universal coverage percentage.
Expect exact target revision/diff and requirements/acceptance when available. Inspect test layout, CI workflows, build/lint/type/format commands, coverage/performance/security thresholds, test data/fixtures, service dependencies, environments, flaky-test policy, and past related failures.
question → focused scenario → matching method → fresh identified artifact → observable/quantitative result → limitations. Capture command, exit status, revision, device/browser/runtime/build variant, environment, data/run count, counts/duration, artifact/log locations, and material side effects. Compare before/after only under comparable conditions; one noisy sample is a lead, not a conclusion.AC-*/risk to actual evidence. Mark verified, failed, not run with reason, or not applicable.security-operations and package/distribution decisions to release-change. Do not let a downstream release controller pre-judge missing quality evidence.Test strategy/evidence matrix, actual commands/results, failure triage, requirement coverage, CI parity, gaps, and quality gate decision. Handoff failures to implementation-execution; successful evidence to documentation-knowledge, review-audit, and release-change.
Relevant tests actually ran; results are tied to the target revision/environment; failures and not-run checks are explicit; acceptance and risk coverage are traceable; residual confidence is calibrated.