Skip to content

pr-review

Thorough code review that understands what your PR is trying to accomplish and checks the implementation against that intent.

Synopsis

/pr-review [target] [base-branch]

/rp1-dev/pr-review [target] [base-branch]

Description

The pr-review command performs comprehensive code review by first understanding what your PR is trying to accomplish (from the description or linked issues), then reviewing each changed file against that intent. Findings are synthesized into an overall assessment with specific, actionable feedback.

Parameters

Parameter Position Required Default Description
TARGET $1 No Current branch PR number, URL, or branch name
BASE_BRANCH $2 No From PR or main Base branch for comparison

Input Resolution

Input Type Example Resolution
Empty - Uses current branch
PR Number 123 Fetches PR metadata via gh
PR URL github.com/.../pull/123 Extracts number, fetches PR
Branch Name feature/auth Uses branch, checks for PR

Architecture

flowchart TB
    subgraph "Phase 0"
        IR[Input Resolution] --> IM[Intent Model]
    end

    subgraph "Phase 1"
        SP[Splitter] --> RU[Review Units]
    end

    subgraph "Phase 2 (Parallel)"
        R1[Sub-Reviewer 1]
        R2[Sub-Reviewer 2]
        RN[Sub-Reviewer N]
        VIS[Visual PR - complex PRs only]
    end

    subgraph "Phase 3"
        SY[Synthesizer] --> JD[Judgment]
    end

    subgraph "Phase 4"
        RP[Reporter] --> MD[Report]
    end

    IM --> SP
    RU --> R1
    RU --> R2
    RU --> RN
    RU -.->|if complex| VIS
    R1 --> SY
    R2 --> SY
    RN --> SY
    VIS -.-> MD
    JD --> RP

Review Dimensions

Each unit is analyzed across 5 dimensions:

Dimension Focus
Correctness Logic errors, edge cases, bugs
Security Vulnerabilities, auth issues
Performance Bottlenecks, inefficiencies
Maintainability Code quality, patterns
Testing Coverage, test quality

Confidence Gating

Findings are filtered by confidence level:

Confidence Action
65%+ Include in report
40-64% (critical/high) Investigate further
Below 40% Exclude from report

Judgment Outcomes

Judgment Meaning
✅ approve Ready to merge
⚠️ request_changes Issues to address
🛑 block Critical problems found

Visual PR Workflow

For complex PRs (significant architectural changes or cross-cutting concerns), the review automatically triggers pr-visual in parallel with sub-reviewers, creating diagrams that help you understand the PR at a glance.

Examples

Review Current Branch

/pr-review

/rp1-dev/pr-review

Review Specific PR

/pr-review 123

/rp1-dev/pr-review 123

Example output: 

✅ PR Review Complete

Judgment: APPROVE
The PR correctly implements user authentication with proper validation.

Findings:
- 🚨 Critical: 0
- ⚠️ High: 1
- 💡 Medium: 3
- ✅ Low: 5

Report: .rp1/work/pr-reviews/pr-123.md

Output

Location: .rp1/work/pr-reviews/<review-id>.md

Contents:

  • Intent summary
  • Judgment with rationale
  • Findings by severity
  • Cross-file issues
  • Recommendations

CI Mode

When running in a CI/CD environment (detected via CI=true environment variable), the command operates in CI mode with additional capabilities. The CI platform is determined by the ci_platform config option (default: github).

CI Mode Behavior

Phase Local Mode CI Mode
P-1: Config Skipped Load .rp1/config/pr-review.yaml, apply env overrides
P0: Input Interactive prompts Extract from GITHUB_EVENT_PATH
P0.5: Visual HTML preview Markdown mode (embedded in summary)
P1-P4: Review Standard Standard
P5: Posting Skipped Post review to GitHub via API

Configuration

Create .rp1/config/pr-review.yaml to enable and configure CI reviews:

enabled: true
verdict: auto
add_comments: true
max_comments: 25
visualize: false

See PR Review Config Reference for all options.

Environment Variables

Override configuration via environment variables:

Variable Effect
RP1_PR_REVIEW_ENABLED Enable/disable review
RP1_PR_REVIEW_VERDICT Set verdict mode
RP1_PR_REVIEW_ADD_COMMENTS Enable/disable inline comments
RP1_PR_REVIEW_VISUALIZE Enable/disable diagrams

Comment Deduplication

In CI mode, the command:

  1. Fetches existing comments from the PR
  2. Compares new findings against existing bot comments
  3. Acknowledges human comments that match findings
  4. Posts only unique findings

This prevents duplicate comments across multiple workflow runs.

GitHub Actions Setup

name: rp1 PR Review

on:
  pull_request:
    types: [opened, synchronize, reopened]

permissions:
  contents: read
  pull-requests: write

jobs:
  review:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0

      - name: Install rp1
        run: curl -fsSL https://rp1.run/install.sh | bash

      - name: Install Claude Code
        run: npm install -g @anthropic-ai/claude-code

      - name: Run PR Review
        env:
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
        run: claude /rp1-dev:pr-review

See Remote PR Review Guide for detailed setup instructions.

See Also