compose-review

Generate policy-aligned PR review JSON from GitHub diffs or local git changes.

1|Updated May 2, 2026
One-click install
npx skills add https://github.com/abeyuya/skills --skill compose-review
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: compose-review
Source: https://github.com/abeyuya/skills/tree/main/plugins/pr-review/skills/compose-review
Command: npx skills add https://github.com/abeyuya/skills --skill compose-review

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

PR diffs and local branch changes often require consistent, policy-driven review comments that are hard to assemble quickly and in the right schema for downstream posting.

Core Features & Use Cases

  • Policy-Driven Review Output: Uses a style reference plus the first matching project instruction file (REVIEW.md / AGENTS.md / .claude/CLAUDE.md / CLAUDE.md) to decide review priorities and formatting.
  • Inline Comment Generation: Produces a comments[] array with normalized file/line targeting and labeled guidance intended for a structured PR-review workflow.
  • Optional External Review Merging (Read-only): Merges findings from one available host/external review skill (preferred order) with the skill’s own review, while avoiding duplicate/irrelevant out-of-diff remarks.

Use Case: For a GitHub PR, generate a complete review body and inline comment set (or a “no issues” response) based on diff content and existing thread/CI context, ready for the orchestrator to post.

Quick Start

Ask the agent to generate a policy-aligned PR review JSON for the current PR diff and return the result as a single un-fenced JSON text.

Frequently Asked Questions about compose-review

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

FAQPage Schema
How do I generate structured inline comments for a GitHub PR review?

You generate structured inline comments by reading PR diffs and producing a single JSON object with a comments array, using normalized file and line targeting for the GitHub PR review workflow.

Can I use local git branch changes to create policy-driven code review output?

Yes, local git branch comparisons can be used to create policy-driven code review output by applying diff-mode selection and referencing project instruction files like REVIEW.md or CLAUDE.md.

What's the best way to enforce a consistent JSON schema for automated PR diff analysis?

The best way to enforce a consistent JSON schema for automated PR diff analysis is to read style references and project instructions, returning a single final JSON object that matches the required schema.

Does this approach support merging external code review findings with my own diff analysis?

Yes, it supports merging findings from one available external host review skill in read-only mode, combining those results with its own review while avoiding duplicate or out-of-diff remarks.

What happens when errors occur during automated code review generation?

When errors occur during code review generation, the process handles them by returning only a simple JSON object containing an error message, ensuring downstream parsing does not break.

Do I need a REVIEW.md file to generate policy-aligned PR review comments?

You do not strictly need a REVIEW.md file; the process searches for the first matching project instruction file among REVIEW.md, AGENTS.md, or CLAUDE.md to decide review priorities.