review

Audits a code diff against its specification and project standards, producing severity-labeled findings and a verdict.

Updated Feb 12, 2026
One-click install
npx skills add https://github.com/kieranpotts/skills --skill review-kieranpotts
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: review
Source: https://github.com/kieranpotts/skills/tree/main/skills/review
Command: npx skills add https://github.com/kieranpotts/skills --skill review-kieranpotts

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Code review is inconsistent and often superficial: reviewers infer intent from the diff, miss untested behavior, and leave ambiguous comments that stall merges. This Skill enforces a disciplined, read-only audit of a proposed change against the specification it claims to satisfy and the project's documented standards, ending in a clear verdict. ## Core Features & Use Cases - Specification-anchored review: Reads the PR description, linked issue, and acceptance criteria before touching code, so findings test the change against stated intent rather than reverse-engineered guesses. - Six-category coverage with severity labels: Systematically checks correctness, design, clarity, test coverage, security, and completeness, labeling every finding Blocking, Suggestion, Nit, or Praise and closing with Approve, Request changes, or Comment. - Non-interactive, read-only operation: Pins an explicit comparison base (e.g., git diff <base>...HEAD), never edits code or configuration, and suits CI or away-from-keyboard workflows. - Use Case: Before pushing a feature branch, ask the agent to review your diff against the issue's acceptance criteria; it returns file-and-line findings grouped by Specification and Standards axes, plus a merge verdict. ## Quick Start Review this branch against main and check it against the acceptance criteria in the linked issue.

Frequently Asked Questions about review

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I review a pull request against its acceptance criteria?▼

Read the PR description and linked issue first to extract the acceptance criteria, then pin a comparison base and read the diff commit by commit. Check each criterion explicitly and quote the specification line for every specification finding.

How to self-review my own diff before opening a PR?▼

Run the full review procedure over your own diff as though it were a stranger's: pin the base with git diff <base>...HEAD, check correctness, tests, and security, then classify findings by severity. Most easy findings surface in self-review before anyone else looks.

What severity labels should code review comments use?▼

Use four labels: Blocking for issues that gate the merge (correctness, security, untested behavior), Suggestion for improvements acceptable as-is, Nit for optional stylistic points, and Praise to reinforce good patterns. Every comment should carry one label so authors know what blocks the merge.

Can an AI code review run non-interactively in CI?▼

Yes, this review workflow runs non-interactively and suits continuous integration. The agent determines parameters from context and environment, only asking where specification or standards artifacts live if nothing settles their location.

When should a code review request changes instead of approving?▼

Request changes when at least one Blocking finding stands, such as a correctness defect, security issue, or new behavior with no test. Approve when nothing blocks, even if Suggestions or Nits remain, since perfection should not hold up delivery.

What are the limitations of automated code review for large diffs?▼

Review effectiveness drops sharply past a few hundred changed lines because reviewers skim and miss issues. For oversized changes, stop and ask the author to split the work into smaller commits rather than attempting a full review.