Review change

Skill kunalsuri/ai-fication-kit/.agents/skills/review-change

AI-Fication Kit: A Simple & Elegant Way for Making any Codebase AI-native through Scaffolded, Human-verified Context and Development Loop.

Install

npx -y skills add kunalsuri/ai-fication-kit --skill review-change

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

2 stars2 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.

What its author says it does

Copied from the file, not written here

Review a completed change against its authorizing spec in fresh context — evidence-based checks, severity-ranked findings, and a written verdict for the human's merge decision. Use when a diff needs reviewing before merge, or when add-feature / fix-bug / implement-spec reaches its review gate.

SKILL.md

2.0 KB, as published. Nobody here has run it

Review a change

The contract: review in a session that did NOT write the change; check with evidence, not assertions; re-run the suites yourself; the written verdict is the deliverable — the human merges, not you.

1. Fresh-context gate

A reviewer sharing the implementer's context inherits the implementer's blind spots. If this session wrote the change, stop and hand the review to a fresh session that did not implement it.

2. Pin the scope

Identify the exact diff (commits / branch / files) and the spec or bugfix doc in ai/lab/specs/ that authorized it. No spec ⇒ that is finding #1, severity blocker: unspecced work.

3. Open the review

Copy ai/lab/reviews/REVIEW_TEMPLATE.md → ai/lab/reviews/REVIEW_<work-id>.md.

4. Check with evidence

For each check in the template — spec conformance, surgical diff, Stability respected, tests, conventions, knowledge updated, provenance clean — record where you looked and what you saw. Re-run the suites the spec names yourself; do not trust the implementer's report.

5. File findings by severity

Any blocker or major ⇒ verdict request-changes; hand the list back to the implementer. Minor / nit findings can ship with notes.

6. Verdict and hand-off

Fill "what the human should double-check" — the judgement calls a mechanical check cannot make. The review itself is [inferred]; the human's merge decision is the real approval, and this document is its evidence.

7. Record

Link the review from the work's row in ai/lab/WORKLOG.md (Review column) and set that row's Status to in-review.