Review

Skill vermaankit2005/shipmate/skills/review

Deep fresh-eyes audit on demand — whole product before shipping, or one sprint on suspicion. Never reopens design; findings get one fix pass.From its SKILL.md

Install

npx -y skills add vermaankit2005/shipmate --skill review

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

1 stars1 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

2.9 KB, as published. Nobody here has run it

/review — the deep audit

Per-sprint review already happens automatically inside /sprint (when the quality gates include review). This command is for the two other moments: the whole-product pass before shipping, and re-examining one sprint because the user's gut says something's off. Typing it is an explicit request — it runs regardless of the gates, manual-mode projects included.

Scope (from the words, not from a question)

"$ARGUMENTS" names a sprint → that sprint's diff, reconstructed from the commit range logged in docs/PLAN.md's sprint log, judged against that sprint's goal and acceptance criteria, plus any user feedback about it.
Blank → the whole product against docs/SCOPE.md: every v1 promise, checked against the running product.

1. Mechanical first (cheap before expensive)

Lint, typecheck, full test suite — report results verbatim. Docs drift: does the README/setup still match reality? Plan drift: ticked boxes vs actual code and commits.

2. Run it

Whole-product mode: start the product per docs/STACK.md, walk the core flows a real user would, AND re-walk every completed sprint's acceptance criteria from docs/PLAN.md — each was verified once, at its own close; this is the only phase that catches later sprints breaking earlier work. Sprint mode: exercise that sprint's acceptance criteria.

3. Fresh eyes

Dispatch ONE shipmate:reviewer agent with the scope (diff or product), the acceptance criteria from docs/PLAN.md / docs/SCOPE.md, and the UI rules from docs/STACK.md. Findings only, ranked, verified.

4. Triage with the user — they're present for this command

Merge everything into one ranked list, present it, then take the verdict with AskUserQuestion (fix all real ones now / queue the non-blockers / decide item by item):

Fix now — correctness bugs, promises not actually kept, dishonest tests. Fix in the main loop, one pass.
Queue — real but not blocking → append to the next sprint's fix-list in docs/PLAN.md (same mechanism as user feedback) so they don't evaporate.
Reject — noise; one line why.

Review is bounded: one review → one fix pass → done. Findings that would change locked decisions (scope, stack, architecture) are never acted on here — they're presented for the user's verdict, because only the user reopens the lock. No re-brainstorming, no loops.

Close

Append a review entry to the sprint log (date, scope, outcome). Commit fixes plainly, no AI co-author trailers. End with one line: what was fixed, what was queued, and — if this was the pre-ship pass — whether the product is honestly ready for the release sprint.