local-review

Reviews pull request diffs for bugs, security issues, and AGENTS.md compliance using a fresh-context subagent.

Updated May 28, 2026
One-click install
npx skills add https://github.com/kma-core/windmill --skill local-review-kma-core
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: local-review
Source: https://github.com/kma-core/windmill/tree/main/.agents/skills/local-review
Command: npx skills add https://github.com/kma-core/windmill --skill local-review-kma-core

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Code reviews done inline in an active session inherit the author's reasoning and rationalizations, causing the reviewer to miss issues that CI catches. This Skill runs the same review locally that GitHub auto-review actions run on PRs, delegating to a fresh-context subagent so findings are unbiased. ## Core Features & Use Cases - Fresh-context review delegation: Spawns a read-only subagent (e.g., Claude Code's branch-diff-reviewer) that reads REVIEW.md policy and AGENTS.md files cold, like CI does. - Structured severity findings: Outputs P0/P1/P2 findings with file paths and line numbers, plus a test coverage assessment. - Optional PR comment posting: With --comment, posts a top-level review or inline line-level comments via the gh CLI and GitHub reviews API. - Use Case: Before pushing a branch, ask for a local review of your diff against main to catch bugs, security issues, and AGENTS.md violations before opening the PR. ## Quick Start Review the current branch diff against main for bugs, security issues, and AGENTS.md compliance.

Frequently Asked Questions about local-review

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

FAQPage Schema
How do I review a PR locally before merging?

Run the local-review skill with a PR number or branch name. It detects the diff against main, delegates the review to a fresh-context subagent that reads REVIEW.md and AGENTS.md, and returns severity-triaged findings with file and line references.

Why does code review need a fresh-context subagent?

A session that iterated on the diff absorbs the author's reasoning and rationalizations, anchoring the reviewer and causing missed issues. A fresh subagent starts cold like CI does, so it surfaces findings the main session would dismiss.

How do I post inline review comments on a GitHub PR?

Pass the --comment flag so the subagent emits a JSON array of line-level comments. The main session then posts them with gh api repos/{owner}/{repo}/pulls/{pr}/reviews using event COMMENT, or uses gh pr review for a top-level comment.

What does the code review output format look like?

Findings are listed as numbered items tagged P0, P1, or P2 with a description and file_path:line_number, followed by a test coverage section. If nothing is found, the output states the change is good to merge.

Can this review run inline in the current session?

No, running inline defeats the purpose because the current session is anchored by prior iteration context. If no fresh-session subagent mechanism is available, the skill instructs you to rerun it in a fresh CLI session.