사용 방법
- 설치 명령을 복사하고 자신의 에이전트 환경에 설치하기 전에 소스 패키지를 확인하세요.
- 필요한 도구, 인증 정보, 지원 환경은 원본 문서에서 확인하세요.
- 입력과 원하는 산출물을 설명하고 출처와 결과를 대조하세요.
npx skills add https://github.com/frostney/known-good-route --skill test-against-spec시작 전 확인
- 설치에는 모델 구독, API 한도, 파일 접근 권한이 포함되지 않습니다.
- 별점 수와 설치 수는 활동 지표이며 독립적인 품질·보안 검증이 아닙니다.
원본 저장소의 제목, 예시, 코드, 표, 링크, 이미지를 유지해 표시합니다.
Test against spec
Establish whether the delivered behavior matches the explicit specification. Report by default. With the exact fix qualifier, fix observed in-scope gaps and retest them. This skill does not replace source review or the repository's full project gate; the caller owns that aggregate gate, and source review and unit-test success alone are not behavior evidence.
Reuse recorded real-interface evidence when the implementation, requirement, environment, and inputs match, except that fix mode reproduces each failure first-hand before fixing it. Run missing or invalidated behavior checks after changes, failures, or unresolved concerns.
Boundaries
- Derive expected behavior only from the user request, issue, confirmed
mini-spec, product documentation, ADR, or another explicit decision. Never infer it from the implementation. Stop for a material conflict or ambiguity.
- Test externally observable outcomes. Use the rendered product for user-facing
behavior and the real API, CLI, library entry point, job, package, migration, or deployed service for other behavior. For qualitative requirements, compare the actual artifact and behavior with the user's reference or acceptance criteria; passing metrics alone cannot establish visual fidelity or usability.
- Do not use implementation source, test source, unit tests, mocks, snapshots,
or a patch as evidence that behavior works. Requirements with no executable behavior are outside this skill's scope and belong to code review or the project gate.
- Prefer a preview deployment when it is available and tied to the exact
revision under test. Use the local environment when no suitable preview exists or when local execution is needed to cover the current working change. Use existing black-box automation when it exercises the real delivered interface and its result is current.
- Do not commit, push, deploy, publish, or change external data unless the user
separately authorized that action. Use disposable test data where possible and ask before a test would create a material external side effect.
Workflow
- Record the specification sources and the exact worktree revision, commit, or
preview revision under test. Separate each externally observable requirement from requirements outside this skill's scope, and select the environment for each behavior under the boundaries above.
- Exercise every testable requirement. Cover the intended path and the most
consequential failure or boundary path. For user-facing changes, interact with the rendered product and cover affected states, accessibility, and relevant viewports.
- Record the environment, setup, action or command, input, expected result, and
observed result. A requirement passes only when the expected outcome is observed against the exact content under test. A submitted check or acknowledgment without a result leaves the requirement unverified.
- When neither a current preview nor the local environment can reproduce
required behavior, report what was attempted and mark the behavior unverified.
- Without
fix, make no edits. Withfix, read
references/fix-mode.md as soon as fix is present, before inspecting implementation source, and follow it for every failed requirement.
Return a structured summary in the active workflow, not a committed or ignored artifact. Include the tested revision, environments, specification sources, each requirement labeled passed, failed, unverified, or out of scope with its evidence or limitation, fixes and changed files when applicable, and whether any behavior remains failed or unverified.




