Test
Verifies a completed build against the spec's acceptance criteria. Only runs when the user explicitly types /test.From its SKILL.md
npx -y skills add bingelp/skills --skill testAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
3.1 KB, 691 tokens by cl100k_base, as published. Nobody here has run it
Test
Overview
Check the implementation against spec.md's acceptance criteria, one by one, with real evidence — not a plausibility argument. This is behavioral verification, not code review: /review follows this step for a final spec-conformance pass, and your existing code-review/security-review tooling covers code quality separately.
When to Use
Explicit invocation only (/test). Requires specs/<slug>/spec.md (for acceptance criteria) and a specs/<slug>/tasks.md with all tasks checked off.
Where artifacts live
Every specs/<slug>/… path below resolves under the repo's shared git dir, not the working tree:
SPECS="$(git rev-parse --path-format=absolute --git-common-dir)/specs" # e.g. …/.git/specs
Storing artifacts there keeps them visible across every session and worktree — including the background-isolated steps Claude Code may switch into automatically — while making them impossible to accidentally commit. Outside a git repo, fall back to ./specs.
Process
- Find
specs/<slug>/spec.mdandspecs/<slug>/tasks.md. If tasks aren't all checked off, stop and tell the user to finish/buildfirst. - For each acceptance criterion in the spec, actually exercise it: run the test suite, run the feature, hit the endpoint, check the UI in a browser — whatever proves the behavior, not just reading the code.
- Append a Verification section to
specs/<slug>/tasks.mdlisting each acceptance criterion by itsAC<n>ID with a pass/fail and the evidence (test name, command output, screenshot description). Cite the ID exactly as written inspec.md; if the spec has a criterion with no ID, flag it rather than inventing one. Cover everyAC<n>in the spec — a missing ID in your Verification section reads as untested. - If something fails, don't paper over it — report it and tell the user to go back to
/build(or/planif the gap is a planning issue). If the failure reveals the spec itself was wrong and anAC<n>should change, that's an upstream edit: change it via/specand reconcile the chain per where/RECONCILE.md — re-plan/re-build/re-test as the change requires. Never edit an AC from inside/testjust to make it pass. - If everything passes, tell the user: "All acceptance criteria verified. Run
/reviewfor a final spec-conformance pass." - If this session is still worktree-isolated when this step ends, ask the user whether to keep or remove the worktree before finishing, per docs/worktrees.md.
Red Flags
- Marking a criterion as passing because the code "looks right" without running anything.
- Testing implementation details instead of the acceptance criteria as written in the spec.
- Quietly weakening an acceptance criterion to make it pass — flag the mismatch to the user instead. A genuinely wrong AC is a
/specchange plus reconciliation (where/RECONCILE.md), not an in-place edit here.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.