PR Review Kit

AI-driven Pull Request code review — structured, multi-perspective, actionable.

Module system, agent YAML, step-file workflows, CLI installer with full IDE integration.

🌐 Website 📖 Full Documentation

Quick Start

# Install into your repo (interactive — recommended)
npx prr-kit install

# Or use the alias
npx pr-review install

# Silent install with all defaults
npx prr-kit install --directory /path/to/repo --modules prr --tools claude-code --yes

Then open your IDE in the installed project and use one of these commands to start:

• /prr-quick — one command, full pipeline (select PR → review → report)
• /prr-master — full menu with all options

Note: The exact command depends on your IDE. See IDE Support for the command specific to your IDE.

How It Works

┌─────────────────────────────────────────────────────────┐
│                    /prr-quick                           │
│              Load config.yaml                           │
└───────────────────────┬─────────────────────────────────┘
                        │
        ╔═══════════════▼════════════════╗
        ║       PHASE 1 — SELECT PR      ║
        ╚═══════════════╤════════════════╝
                        │
             ┌──────────▼──────────┐
             │ 1a. git fetch origin │
             └──────────┬──────────┘
                        │
             ┌──────────▼───────────────┐
             │ 1b. List open PRs/MRs    │
             │     + recent branches    │
             └──────────┬───────────────┘
                        │
             ┌──────────▼───────────────┐
             │ 1c. ⌨️  Select PR/branch  │  ← USER INPUT
             │     Enter PR# or name    │
             └──────────┬───────────────┘
                        │
             ┌──────────▼──────────┐
             │ 1d. Load diff        │
             │     count files/lines│
             └──────────┬──────────┘
                        │
             ┌──────────▼──────────────────────────┐
             │ 1e. Create session folder            │
             │     {output}/{date}-{slug}/          │
             └──────────┬──────────────────────────┘
                        │
             ┌──────────▼──────────────────────────┐
             │ 1f. Generate diffs/ folder           │
             │     one .md per changed file         │
             └──────────┬──────────────────────────┘
                        │
        ╔═══════════════▼════════════════╗
        ║      PHASE 2 — DESCRIBE PR     ║
        ╚═══════════════╤════════════════╝
                        │
             ┌──────────▼──────────────┐
             │ 2a. Classify PR type     │
             │     bugfix/feature/...   │
             └──────────┬──────────────┘
                        │
             ┌──────────▼──────────────┐
             │ 2b. File-by-file         │
             │     walkthrough          │
             └──────────┬──────────────┘
                        │
             ┌──────────▼──────────────┐
             │ 2c. Print PR description │
             └──────────┬──────────────┘
                        │
        ╔═══════════════▼═══════════════════════╗
        ║   PHASE 2.5 — COLLECT PR CONTEXT      ║
        ╚═══════════════╤═══════════════════════╝
                        │
             ┌──────────▼──────────────────────────┐
             │ Step 1. Analyze changed files        │
             │  detect stacks, domains, file types  │
             └──────────┬──────────────────────────┘
                        │
             ┌──────────▼──────────────────────────┐
             │ Step 2. Collect context from:        │
             │  CLAUDE.md · CONTRIBUTING.md         │
             │  .eslintrc · tsconfig · pyproject    │
             │  ARCHITECTURE.md · docs/**           │
             │  @context/@security annotations      │
             │  stack rules (vue3/react/django/...)  │
             │  MCP tools · RAG · URL sources       │
             └──────────┬──────────────────────────┘
                        │
             ┌──────────▼──────────────────────────┐
             │ Step 3. ⌨️  User instructions        │  ← USER INPUT
             │  scope / focus / requirements /      │
             │  context — or Enter for full review  │
             └──────────┬──────────────────────────┘
                        │
               ┌────────┴────────┐
               ▼                 ▼
          [provided]         [empty]
        parse scope &      scope = "all"
        focus/req/ctx      full standard
               │                 │
               └────────┬────────┘
                        │
             ┌──────────▼──────────────────────────┐
             │ Step 4. Build pr-context.yaml        │
             │  all collected context + user_instr  │
             │  → {session_output}/pr-context.yaml  │
             └──────────┬──────────────────────────┘
                        │
        ╔═══════════════▼════════════════╗
        ║       PHASE 3 — REVIEW         ║
        ╚═══════════════╤════════════════╝
                        │
             ┌──────────▼──────────────────────────┐
             │ Scope gate                           │
             │ read user_instructions.review_scope  │
             └──────┬───────────────────────────────┘
                    │
         ┌──────────┴──────────┐
    [in scope]           [not in scope]
         │                     │
         ▼                     ▼
  ┌─────────────┐        ⏭️ skipped
  │  GR · SR    │        (no output file)
  │  PR · AR    │
  │     BR      │
  └──────┬──────┘
         │
        ╔▼════════════════════════╗
        ║  PHASE 4 — REPORT       ║
        ╚╤════════════════════════╝
         │
         │  Compile findings → sort by severity
         │  🔴 Blockers → 🟡 Warnings
         │  🟢 Suggestions → ❓ Questions
         │  Write → final-review.md
         │
        ╔▼════════════════════════╗
        ║  PHASE 5 — DONE         ║
        ╚╤════════════════════════╝
         │
         ├── auto_post_comment: true  → post inline comments automatically
         └── auto_post_comment: false → ⌨️  type PC to post, Enter to finish

The framework installs into your project as a _prr/ folder. Agents and workflows are Markdown/YAML files that your AI IDE reads and executes — no server, no background process, no API keys required beyond your IDE's AI.

Configuration

The installer handles configuration interactively — no manual file editing required. During npx prr-kit install, you'll be prompted for your name, language, and output folder. Platform, repo, and target directory are auto-detected from your git remote.

All values are written to _prr/prr/config.yaml. Full schema overview:

# ─── Identity ──────────────────────────────────────────────────────────────
user_name: YourName                    # Your name — used in review reports
communication_language: English        # Any language: English | Vietnamese | Japanese | French | …

# ─── Project ───────────────────────────────────────────────────────────────
project_name: my-project               # Display name in reports (cosmetic only)

# ─── Output ────────────────────────────────────────────────────────────────
review_output: ./_prr-output/reviews   # Where review reports + context files are written
auto_post_comment: false               # true → auto-post findings after every review (skips PC prompt)

# ─── Context Collection ────────────────────────────────────────────────────
context_collection:
  enabled: true                        # false → disable context collection entirely
  mode: pr-specific                    # only value: pr-specific (always fresh, never cached)

  # Sources below are auto-detected — override only if needed:
  # primary_sources:  [CLAUDE.md, AGENTS.md, .github/CLAUDE_CODE_RULES.md, .clauderules]
  # config_files:     [.eslintrc*, .prettierrc*, tsconfig.json, vite.config.*, webpack.config.*, …]
  # standards_docs:   [CONTRIBUTING.md, ARCHITECTURE.md, docs/**/*.md]
  # inline_annotations: { enabled: true, patterns: [@context:, @security:, @pattern:, @rule:] }

# ─── External Sources ──────────────────────────────────────────────────────
# MCP tools + RAG systems available in your AI IDE session.
# Agent auto-discovers tools and maps them to declared intents.
external_sources:
  enabled: false                       # true → activate MCP + RAG enrichment

  mcp:
    enabled: true                      # toggle MCP independently of master switch
    intents:                           # what kinds of context to fetch via MCP tools
      - knowledge_base                 # Confluence, Notion → team standards, ADRs
      - project_management             # Jira, Linear → linked issue + acceptance criteria
      - design                         # Figma, Zeplin → design specs (UI PRs only)
      # - code_intelligence            # Sourcegraph → similar patterns
    hints:
      branch_issue_pattern: "([A-Z]+-\\d+)"  # regex to extract issue key from branch name
                                              # e.g. feature/ENG-123-auth → ENG-123

  rag:
    enabled: false                     # true → query RAG systems (vector DB, embeddings)
    intents:
      - similar_patterns               # find similar code in the codebase
      - past_decisions                 # previous review decisions for similar code
      # - architecture_examples        # embedded architecture docs

  sources: []                          # plain URL sources — always fetched via WebFetch
  # sources:
  #   - type: url
  #     name: Shared ESLint config
  #     url: https://raw.githubusercontent.com/org/standards/main/eslint.md
  #   - type: url
  #     name: Security guidelines
  #     url: https://wiki.company.com/public/security-standards

See CONFIGURATION.md for detailed explanations, examples, and FAQs.

Platform Support

| Feature | None / Local | GitHub | GitLab | Azure DevOps | Bitbucket | |---------|:---:|:---:|:---:|:---:|:---:| | Core review (diff analysis) | ✅ | ✅ | ✅ | ✅ | ✅ | | Auto-detect platform from remote URL | — | ✅ | ✅ | ✅ | ✅ | | List open PRs/MRs | ❌ | ✅ gh | ✅ glab | ✅ az | ✅ bb | | Select PR by number (auto base/head) | ❌ | ✅ | ✅ | ✅ | ✅ | | Load diff via platform CLI | ❌ | ✅ gh pr diff | ✅ glab mr diff | ⚠️ git fallback | ⚠️ git fallback | | Post inline code comments | ❌ | ✅ Reviews API | ✅ MR discussions | ✅ PR threads | ✅ REST API | | Post summary review | ❌ | ✅ | ✅ | ✅ | ✅ | | Review verdict (approve / request changes) | ❌ | ✅ | ✅ | ✅ | ❌ | | Required CLI | — | gh | glab | az + extension | bb / curl |

None / Local mode: all review analysis runs locally via git diff — no platform CLI required. Findings are saved to _prr-output/reviews/ only.

Review Workflow

Quick mode — one command, full pipeline

/prr-quick

Runs automatically: select PR → describe → collect context → review → report Pauses twice for user input: once to select the PR/branch, once for review instructions (scope, focus, requirements — or Enter for a full standard review).

Manual mode — step by step

| Code | Command | Description | |------|---------|-------------| | SP | Select PR | Fetch latest → list open PRs or branches → select head + base → load diff | | DP | Describe PR | Classify PR type, generate summary, file-by-file walkthrough | | CC | Collect Context | Build PR-specific knowledge base from project docs, config files, standards, and external tools — run after DP, before any review | | GR | General Review | Logic, naming, readability, DRY, best practices, etc. — adapted to your stack | | SR | Security Review | OWASP Top 10, secrets, auth, rate limits, injection, etc. — adapted to your project | | PR | Performance Review | N+1 queries, memory leaks, async patterns, caching, etc. — adapted to your stack | | AR | Architecture Review | SOLID, layers, coupling, consistency with codebase, etc. — adapted to your architecture | | BR | Business Review | User impact, business risk, feature completeness, data safety, observability — adapted to your project | | PM | Party Mode 🎉 | All reviewers discuss the PR together in one collaborative session | | IC | Improve Code | Concrete BEFORE/AFTER code suggestions | | AK | Ask Code | Q&A about specific changes in this PR | | RR | Generate Report | Compile all findings → Markdown report in _prr-output/reviews/ | | PC | Post Comments | Post inline review comments to GitHub, GitLab, Azure DevOps, or Bitbucket PR | | SS | Select Session | List past review sessions and resume one | | CL | Clear | Remove context files and/or review reports from output folder | | HH | Help | Show this guide |

Selecting a PR (SP step)

When platform is detected — lists open PRs/MRs via platform CLI:

#45  "Add OAuth2 login"      feature/oauth → main    @alice  3h ago
#44  "Fix memory leak"       fix/memory    → main    @bob    1d ago

Enter PR number → base and head resolved automatically.

When no platform detected (local mode) — asks explicitly for both branches:

🎯 Head branch (the branch to review)?
   • Enter a number from the list  (e.g., 1)
   • Type the branch name directly  (e.g., feature/my-feature)

🎯 Base branch (what to diff against)?
   • Press Enter for default [main]
   • Type the branch name directly  (e.g., develop)

Review Agents

| Agent | Slash Command | Speciality | |-------|--------------|------------| | PRR Master | /prr-master | Orchestrator — routes all workflows, full menu | | PRR Quick | /prr-quick | One-command full pipeline (select → review → report) |

Specialist reviewer agents are orchestrated internally by the master agent and party-mode workflow. Use [PM] Party Mode from the master menu (/prr-master) to run all reviewers together in one collaborative session.

Reviewers at a Glance

| Reviewer | Focus | Key questions | |---|---|---| | 👁️ General (GR) | Code quality + stack practices | Is the logic correct? Naming clear? DRY? Tests present? (adapted to your stack) | | 🔒 Security (SR) | OWASP Top 10 + stack threats | XSS? Injection? Secrets exposed? Auth correct? (adapted to your stack) | | ⚡ Performance (PR) | Efficiency + stack patterns | N+1 queries? Memory leaks? Missing await? (adapted to your stack) | | 🏗️ Architecture (AR) | Structure + conventions | Layer violations? Coupling? Consistent with codebase? (adapted to your architecture) | | 💼 Business (BR) | Real-world impact | User impact? Business risk? Feature completeness? Data safe? Observability? (adapted to your project) |

Checks are adaptive — each reviewer skips categories not relevant to your project and generates additional checks based on detected stacks, project guidelines, and inline annotations.

Business Review (BR) runs last and translates technical findings into business language — user impact, GDPR risk, migration safety, deployment recommendations, and post-ship monitoring checklist.

Severity Levels

All findings use a standard format:

• 🔴 [BLOCKER] — Must fix before merge
• 🟡 [WARNING] — Should fix (with explanation)
• 🟢 [SUGGESTION] — Nice-to-have improvement
• ❓ [QUESTION] — Needs clarification from author

Context Collection

In [QR] Quick Review: context is collected automatically as Phase 2.5 — no manual step needed. In manual mode (SP → DP → reviews): run [CC] Collect Context after [DP] and before starting any review.

Either way, the agent pauses once for your instructions before building the knowledge base:

1. Analyzes changed files to detect domains (authentication, state-management, etc.)
2. Reads relevant config files (.eslintrc, .prettierrc, tsconfig.json) and standards docs (CONTRIBUTING.md, ARCHITECTURE.md)
3. Extracts inline @context: / @security: / @pattern: annotations from the diff
4. Optionally queries MCP tools (Confluence, Jira, Figma) and RAG systems if configured
5. Asks for review instructions — scope (only security), focus, requirements, or context. Press Enter for a full standard review
6. Writes pr-context.yaml inside the session folder — loaded by all reviewers

See CONFIGURATION.md for MCP intents, RAG setup, and URL sources.

Inline Code Comments

When [PC] Post Comments is run, it posts findings as inline code comments on the exact file and line — the same experience as a human reviewer. Platform and repo are auto-detected from your git remote.

| Platform | Method | Required CLI | |----------|--------|-------------| | GitHub | Reviews API | gh auth login | | GitLab | MR Discussions API | glab auth login | | Azure DevOps | PR Threads API | az login | | Bitbucket | Inline Comments REST API | bb / curl |

Supported IDEs

Antigravity, Augment Code, Claude Code, Cline, Codex, Crush, Cursor, Gemini CLI, GitHub Copilot, iFlow, Kilo, Kiro, OpenCode, QwenCoder, Roo Code, Rovo Dev, Trae, Windsurf

Requirements

• Node.js 20+
• Git
• Platform CLI (optional — only needed for PR listing and inline comments):
  • GitHub: gh
  • GitLab: glab
  • Azure DevOps: az + Azure DevOps extension
  • Bitbucket: bb or curl

Development

npm install
npm test

Project Structure

prr-kit/
├── src/
│   ├── core/
│   │   ├── agents/
│   │   │   └── prr-master.agent.yaml   # Master orchestrator + menu
│   │   ├── tasks/
│   │   │   ├── help.md                 # [HH] Help
│   │   │   ├── clear.md                # [CL] Clear output files
│   │   │   └── workflow.xml            # Workflow engine rules
│   │   └── workflows/
│   │       └── party-mode/             # [PM] Party Mode — all reviewers together
│   └── prr/
│       ├── agents/                     # Specialist reviewer agents (GR SR PR AR)
│       ├── config-template.yaml        # Full config template with all options
│       └── workflows/
│           ├── 1-discover/             # [SP] Select PR
│           ├── 2-analyze/
│           │   ├── describe-pr/        # [DP] Describe PR
│           │   └── collect-pr-context/ # Auto: build PR-specific knowledge base
│           ├── 3-review/
│           │   ├── general-review/     # [GR] Logic, naming, DRY
│           │   ├── security-review/    # [SR] OWASP, secrets, auth
│           │   ├── performance-review/ # [PR] N+1, async, memory
│           │   ├── architecture-review/# [AR] SOLID, layers, coupling
│           │   └── business-review/    # [BR] User impact, risk, completeness
│           ├── 4-improve/              # [IC] Improve Code
│           ├── 5-ask/                  # [AK] Ask Code
│           ├── 6-report/               # [RR] Report + [PC] Post Comments
│           └── quick/                  # [QR] Full pipeline in one command
├── tools/
│   └── cli/                            # CLI installer + IDE handlers
├── test/                               # Schema + component tests
├── CONFIGURATION.md                    # Full config reference (MCP, RAG, context)
└── docs/
    └── assets/
        ├── banner.svg
        └── how-it-works.svg

Support the Project

PR Review Kit is free for everyone — and always will be. If you'd like to support development:

• ⭐ Star on GitHub — Click the star icon near the top right of this page
• ☕ Buy Me a Coffee — Fuel the development

Contributing

Contributions are welcome! See CONTRIBUTING.md for guidelines on adding workflows, reviewer agents, IDE integrations, and more.

License

MIT © mrquangthai278