review-spec — Claude Code skill: check code against feature specs
review-spec is a Claude Code skill that compares code with feature docs and reports mismatches in both directions. Run it mid-feature or after completion.
no license ★ 155 updated 2026-10-29
npx skills add serpro69/claude-toolbox --skill review-spec -a claude-codeWhat review-spec does
The skill opens the feature folder at /docs/feat/wip/[feature]/ with design.md, implementation.md and tasks.md inside, then goes through the code looking for drift. Run it mid-feature to check finished tasks only, or after the whole feature is done. Mismatches surface in both directions: code that left the spec behind, and specs that no longer match the code.
Every finding carries a type (MISSING_IMPL, EXTRA_IMPL, SPEC_DEV, DOC_INCON, OUTDATED_DOC, AMBIGUOUS), a P0–P3 severity, and a confidence score from 1 to 10 with the reasoning behind it. Standard mode $kk:review-spec makes one pass in the current context. Isolated mode $kk:review-spec:isolated hands detection to a spec-reviewer sub-agent that never touched the code; it costs more, and a failed sub-agent sends you back to standard mode. Under an IaC profile such as Kubernetes or Terraform, the manifests count as the implementation.
None of this works without the docs in place: the three files must already sit in /docs/feat/wip/[feature]/. Before looking at code the skill loads the repo's convention files, shared-capy-knowledge-protocol.md and shared-profile-detection.md, and it keeps a fixed order — spec first, code after. Codex works too.
What it helps with
- catch a requirement from
design.mdthat never made it into the code - spot a caching layer the specs never asked for
- check manifest values against the design in a Terraform profile
- tell a stale doc apart from a real deviation
- get a confidence score on every finding before merge
Example prompts for the agent
- Review the feature in
/docs/feat/wip/auth/against its design and implementation docs, run$kk:review-spec. - Run
$kk:review-spec:isolated— I want a reviewer that didn't write this code. - Check only the finished tasks in
tasks.mdwhile the feature is still in progress.
How to install review-spec
1Skills CLI (Vercel)
npx skills add serpro69/claude-toolbox --skill review-spec -a claude-code2Manually: put the files in place
~/.claude/skills/review-spec/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/review-spec/
1Skills CLI (Vercel)
npx skills add serpro69/claude-toolbox --skill review-spec -a codex2Manually: put the files in place
.agents/skills/review-spec/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add serpro69/claude-toolbox --skill review-spec -a cursor2Manually: put the files in place
.cursor/skills/review-spec/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install serpro69/claude-toolbox kodex-plugin/skills/review-spec2Manually: put the files in place
.github/skills/review-spec/SKILL.md1Skills CLI (Vercel)
npx skills add serpro69/claude-toolbox --skill review-spec -a gemini-cli2Platform CLI
gemini skills install https://github.com/serpro69/claude-toolbox --path kodex-plugin/skills/review-specManually: put the files in place
.devin/skills/review-spec/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about review-spec
How do I install review-spec in Claude Code?
Run npx skills add serpro69/claude-toolbox --skill review-spec -a claude-code; the skill ends up in ~/.claude/skills/review-spec/SKILL.md.
What does review-spec do?
It compares the implementation with the feature's design.md, implementation.md and tasks.md, flags missing, extra or stale pieces, and tags each finding with a type, a P0–P3 severity and a confidence score.
Why use isolated mode?
Detection goes to a separate spec-reviewer sub-agent that did not write the code. It costs more than a standard pass, and the skill falls back to standard mode if that sub-agent fails.