pr-review

Reviews pull requests against intent and publishes verdict-mapped GitHub reviews with prioritized findings.

Updated Jul 8, 2026
One-click install
npx skills add https://github.com/Zhachory1/autopraxis --skill pr-review-zhachory1
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: pr-review
Source: https://github.com/Zhachory1/autopraxis/tree/main/skills/pr-review
Command: npx skills add https://github.com/Zhachory1/autopraxis --skill pr-review-zhachory1

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Code review often drifts into nitpicking without grounding in the actual intent of a change, and feedback gets lost, duplicated, or posted against stale commits. This Skill anchors PR review to tickets, design docs, and acceptance criteria, then delivers evidence-backed, prioritized findings directly to GitHub. ## Core Features & Use Cases - Intent-anchored review: Gathers tickets, PRDs, and author notes before judging code, checking architecture and boundaries before line-level issues. - GitHub review publishing: Maps verdicts to APPROVE, REQUEST_CHANGES, or COMMENT events with idempotency markers, head-SHA pinning, and duplicate-feedback prevention via a prior-feedback ledger. - Bounded delta re-review: After author revisions, re-reviews only changed files and unresolved findings, with escalation when blockers persist. - Use Case: A developer pastes a PR URL and asks for a review; the agent gathers context, checks architecture and correctness, runs focused tests, and posts a single pinned GitHub review with inline findings and a merge recommendation. ## Quick Start Review the pull request at the given GitHub URL and post a verdict-appropriate review with prioritized findings.

Frequently Asked Questions about pr-review

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

FAQPage Schema
How do I review a GitHub pull request with an AI agent?▼

Provide the PR URL and any linked tickets or design docs. The agent gathers intent and prior feedback, checks architecture and correctness, then publishes a GitHub review mapped to its verdict: APPROVE, REQUEST_CHANGES, or COMMENT.

How to avoid duplicate comments when re-reviewing a pull request?▼

Build a prior-feedback ledger from all paginated issue comments, reviews, and inline threads before drafting findings. Skip concerns already raised unless new code or material evidence changes the area, and cite the earlier feedback when revisiting.

Can an AI agent approve or request changes on a GitHub PR?▼

Yes, it can submit APPROVE, REQUEST_CHANGES, or COMMENT reviews based on its evidence-backed verdict, but it never merges. Merge decisions remain with a human through an explicit approval gate.

What happens if the PR head changes before the review is posted?▼

The agent re-fetches the head SHA immediately before publishing. If it changed, it re-reviews the delta first, then submits the review pinned with commit_id to the verified SHA and reports if the posted review was superseded.

When should a pull request review escalate to a council or deeper mode?▼

Escalate for high-risk architecture, security, privacy, reliability, ML/statistical changes, data migrations, or conflicting reviewer judgments. Ordinary PRs use no council; a single domain concern uses one lens only.