skills.rest
chris-prener
chris-prenerCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

code-review

Reviews git diffs via an independent subagent and produces structured findings reports.

Updated Jul 29, 2026
Search Tags:#code review#pull request#github#subagent#git diff#findings report#pre-pr gate
One-click install
npx skills add https://github.com/chris-prener/dev-kit --skill code-review-chris-prener
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: code-review
Source: https://github.com/chris-prener/dev-kit/tree/main/dev-kit/skills/code-review
Command: npx skills add https://github.com/chris-prener/dev-kit --skill code-review-chris-prener

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? In a solo-plus-AI workflow there is no human PR reviewer, so bugs, design flaws, and silent architectural decisions can slip into pull requests unchecked. This Skill standardizes a rubber-duck-style AI review pass that closes that gap before a PR is created. ## Core Features & Use Cases - Independent subagent review: Spawns a fresh subagent with no stake in the implementation to review the diff against issue acceptance criteria, ADRs, and repo invariants. - Severity-tagged findings with auto-filing: Categorizes findings as BLOCKER/HIGH/MEDIUM/NIT and auto-files non-trivial ones as GitHub issues with dedup markers. - Gate and self-review modes: Runs as the mandatory pre-PR gate invoked by pr-orchestrator, or ad hoc as a self-review before pushing, with a minimal whitelist short-circuit for trivial doc-only changes. - Use Case: Before opening a PR that closes issue #42, run the review gate to get a structured report at .github/audit-reports/ with a CLEAN, ADVISORY, or BLOCKED verdict and auto-filed issues for each real finding. ## Quick Start Ask the AI to review the diff on the current branch against the base branch before creating the pull request.

Frequently Asked Questions about code-review

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

FAQPage Schema
How do I run an AI code review on a git diff before creating a PR?▼

Invoke the review mode on your branch, which computes the unified diff against the base branch and spawns an independent subagent to analyze it. Findings are written to a report in .github/audit-reports/ with a CLEAN, ADVISORY, or BLOCKED verdict.

How does automated code review categorize findings by severity?▼

Findings are tagged BLOCKER for bugs and invariant violations, HIGH for design flaws and missing tests on non-trivial changes, MEDIUM for doc staleness and misleading naming, and NIT/QUESTION for informational items. Style and formatting findings are forbidden since lint covers those.

Can AI code review auto-file GitHub issues for findings?▼

Yes, each BLOCKER, HIGH, and MEDIUM finding is auto-filed as a GitHub issue through the backlog skill with an autofile-id marker for deduplication. NIT and QUESTION findings appear only in the report and file no issues.

Does the review gate run on small documentation-only changes?▼

A minimal mode short-circuits to a quick pass when the diff is 20 lines or fewer and touches only whitelisted files like CHANGELOG.md, README.md, or top-level docs. It requires explicit operator confirmation, and any source, config, or CI change always gets full review.

What are the limitations of AI-based pre-PR code review?▼

The skill is read-only and single-repo: it does not merge PRs, post comments to the PR conversation, auto-fix findings, or handle style and lint concerns. Review depth is also reduced when the PR does not close an issue with acceptance criteria.

Related Skills

Explore More Software Engineering Skills→
genlayerlabsgenlayerlabs
157

create-pr

Create GitHub pull requests from branch diffs and commit logs.

Official
Intermediate
MaelwalserMaelwalser

project-flow-ops

Triage GitHub issues and pull requests into Linear execution flows.

Community
Intermediate
MrJmpl3MrJmpl3

0129-code-review-excellence

Guide pull request reviews for logic, security, performance, and maintainability issues.

Community
Intermediate
josipmusajosipmusa
13

jds-execute

Execute confirmed implementation plans by dispatching isolated subagents with rubber-duck verification.

Community
Advanced