Skip to content

VitaDynamics/kimi-actions

BranchesTags

Repository files navigation

Kimi Code Review Action

🌗 AI-powered code review using Kimi (Moonshot AI)

Architecture

┌─────────────────────────────────────────────────────────────────────┐
│                              GitHub                                 │
│  ┌──────────────────────────────────────────────────────────────┐   │
│  │  PR Events  │  PR Comments  │  Inline Comments               │   │
│  │             │  /review      │  /ask                          │   │
│  │             │  /ask         │                                │   │
│  └──────────────────────────────────────────────────────────────┘   │
└─────────────────────────────────────┬───────────────────────────────┘
                                      │
                                      ▼
┌────────────────────────────────────────────────────────────────────┐
│                    GitHub Actions (Docker)                         │
├────────────────────────────────────────────────────────────────────┤
│                                                                    │
│  ┌───────────────────────────────────────────────────────────────┐ │
│  │                      main.py                                  │ │
│  │  Event Router: PR events → /review, /ask commands             │ │
│  └───────────────────────────────────────────────────────────────┘ │
│                              │                                     │
│                              ▼                                     │
│  ┌───────────────────────────────────────────────────────────────┐ │
│  │                    Tools Layer                                │ │
│  │                                                               │ │
│  │  ┌──────────┐              ┌──────────┐                       │ │
│  │  │ Reviewer │              │   Ask    │                       │ │
│  │  │ /review  │              │  /ask    │                       │ │
│  │  └────┬─────┘              └────┬─────┘                       │ │
│  │       └──────────┬──────────────┘                             │ │
│  │                  ▼                                            │ │
│  │         ┌────────────────┐                                    │ │
│  │         │    BaseTool    │                                    │ │
│  │         │  • clone_repo  │                                    │ │
│  │         │  • run_agent   │                                    │ │
│  │         │  • get_skill   │                                    │ │
│  │         └────────────────┘                                    │ │
│  └───────────────────────────────────────────────────────────────┘ │
│                              │                                     │
│                              ▼                                     │
│  ┌───────────────────────────────────────────────────────────────┐ │
│  │                   SkillManager                                │ │
│  │  Load SKILL.md and set skills_dir for Agent SDK               │ │
│  └───────────────────────────────────────────────────────────────┘ │
│                              │                                     │
│         ┌────────────────────┴────────────────────┐                │
│         ▼                                         ▼                │
│  ┌──────────────────┐                   ┌──────────────────┐       │
│  │  Kimi Agent SDK  │                   │   GitHub API     │       │
│  │   (kimi-k2.5)    │                   │     (REST)       │       │
│  │                  │                   │                  │       │
│  │ • Auto token mgmt│                   │ • Get PR diff    │       │
│  │ • Script exec    │                   │ • Post comments  │       │
│  │ • Context mgmt   │                   │ • Get PR info    │       │
│  │ • Markdown output│                   │                  │       │
│  └──────────────────┘                   └──────────────────┘       │
│                                                                    │
└────────────────────────────────────────────────────────────────────┘

Features

  • 🔍 /review - Comprehensive code review of all PR changes
  • 💬 /ask - Interactive Q&A about the PR or specific code
  • 🧠 Agent Skills - Modular capability extension with custom review rules
  • 🌐 Multi-language support (English/Chinese)
  • ⚙️ Configurable review strictness
  • 🎯 Direct Markdown Output - Clean, readable reviews powered by Agent SDK
  • 🚀 Simplified Architecture - Agent SDK handles all context and token management

Quick Start

1. Get Kimi API Key

  1. Visit Moonshot AI Platform
  2. Register/Login
  3. Go to "API Key Management"
  4. Click "Create API Key"
  5. Copy the generated API Key

2. Configure GitHub Secrets

  1. Go to your GitHub repository
  2. Click Settings → Secrets and variables → Actions
  3. Click New repository secret
  4. Add KIMI_API_KEY with the API Key from step 1
  5. (Optional) Add KIMI_BASE_URL if using a custom API endpoint (defaults to https://api.moonshot.cn/v1)

3. Create Workflow File

# .github/workflows/kimi-review.yml
name: Kimi Code Review

on:
  pull_request:
    types: [opened, synchronize, reopened]
  issue_comment:
    types: [created]
  pull_request_review_comment:
    types: [created]

permissions:
  contents: read
  pull-requests: write

jobs:
  kimi-review:
    runs-on: ubuntu-latest
    if: |
      github.event_name == 'pull_request' ||
      (github.event_name == 'issue_comment' &&
       github.event.issue.pull_request &&
       startsWith(github.event.comment.body, '/')) ||
      (github.event_name == 'pull_request_review_comment' &&
       startsWith(github.event.comment.body, '/'))
    steps:
      - name: Get PR ref (for comments)
        id: get-pr
        if: github.event_name == 'issue_comment' || github.event_name == 'pull_request_review_comment'
        uses: actions/github-script@v7
        with:
          script: |
            const prNumber = context.issue?.number || context.payload.pull_request?.number;
            const pr = await github.rest.pulls.get({
              owner: context.repo.owner,
              repo: context.repo.repo,
              pull_number: prNumber
            });
            core.setOutput('ref', pr.data.head.ref);
            core.setOutput('sha', pr.data.head.sha);

      - uses: actions/checkout@v4
        with:
          ref: ${{ (github.event_name == 'issue_comment' || github.event_name == 'pull_request_review_comment') && steps.get-pr.outputs.ref || github.head_ref }}

      - uses: xiaoju111a/kimi-actions@main
        with:
          kimi_api_key: ${{ secrets.KIMI_API_KEY }}
          kimi_base_url: ${{ secrets.KIMI_BASE_URL }}  # Optional
          github_token: ${{ secrets.GITHUB_TOKEN }}
          auto_review: 'false'  # Use /review command instead

Commands

PR Commands

Use these commands in PR comments:

Command Description Usage Location
/review Comprehensive code review of all PR changes PR comment area
/ask <question> Q&A about the PR or specific code PR comment area or Files changed tab (inline)
/help Show help message PR comment area

💡 Using /ask for code-specific questions:

  • In PR comment area: Ask general questions about the entire PR
  • In Files changed tab: Click the + button next to a line of code, then use /ask <question> to ask about that specific code

🔄 Avoiding Duplicate Reviews:

  • The bot tracks the last reviewed commit SHA
  • If you run /review again without new commits, it will show "✅ No new changes since last review"
  • This prevents wasting tokens on unchanged code

Configuration

Action Inputs

- uses: xiaoju111a/kimi-actions@main
  with:
    # Required
    kimi_api_key: ${{ secrets.KIMI_API_KEY }}
    github_token: ${{ secrets.GITHUB_TOKEN }}
    
    # Optional
    kimi_base_url: ${{ secrets.KIMI_BASE_URL }}  # Custom API endpoint (optional, defaults to https://api.moonshot.cn/v1)
    language: 'en-US'               # Response language: zh-CN, en-US
    model: 'kimi-k2.5'              # Kimi model (default: kimi-k2.5)
    review_level: 'normal'          # Review strictness: strict, normal, gentle
    max_files: '50'                 # Max files to review
    exclude_patterns: '*.lock,*.min.js'  # File patterns to exclude
    auto_review: 'false'            # Auto review on PR open (default: false, use /review command instead)

Repository Config (.kimi-config.yml)

Create .kimi-config.yml in your repo root to customize behavior:

# Category toggles
categories:
  bug: true
  performance: true
  security: true

# Replace built-in skills with custom ones
skill_overrides:
  code-review: my-company-review

# Ignore files
ignore_files:
  - "*.test.ts"
  - "**/__mocks__/**"

# Extra instructions
extra_instructions: |
  Focus on security issues.

Custom Skills (Claude Skills Standard)

Create .kimi/skills/ directory in your repo, each skill is a folder:

.kimi/skills/
├── react-review/
│   ├── SKILL.md           # Required: core instructions
│   ├── scripts/           # Optional: executable scripts
│   │   └── check_hooks.py
│   └── references/        # Optional: reference documents
│       └── hooks-rules.md
└── company-rules/
    └── SKILL.md

SKILL.md format:

---
name: react-review
description: React code review expert
triggers:
  - react
  - jsx
  - hooks
---

# React Review Focus

## Hooks Rules
- Hooks can only be called at the top level of function components
- Cannot call Hooks inside conditionals

## Performance
- Check if useMemo/useCallback is needed

Skills are automatically triggered based on PR code content.

Models

Model Context Notes
kimi-k2.5 256K Default, latest model with best performance
kimi-k2-thinking-turbo 256K Faster thinking model
kimi-k2-thinking 256K More thorough reasoning, slower

All commands use Kimi Agent SDK with kimi-k2.5 model by default.

The Agent SDK automatically handles large PRs with its 256K context window.

Review Categories

Category Description Examples
Bug Code defects Unhandled exceptions, null pointers, logic errors
Security Security vulnerabilities SQL injection, XSS, auth flaws
Performance Performance issues O(n²) algorithms, N+1 queries

Project Structure

kimi-actions/
├── action.yml                  # GitHub Action definition
├── Dockerfile                  # Docker container config
├── requirements.txt            # Python dependencies
├── tests/                      # Unit tests (115 tests)
└── src/
    ├── main.py                 # Entry point, event routing
    ├── action_config.py        # Action config (env vars)
    ├── repo_config.py          # Repo config (.kimi-config.yml)
    ├── github_client.py        # GitHub API client
    ├── skill_loader.py         # Skill loading/management
    ├── tools/                  # Command implementations (Agent SDK)
    │   ├── base.py             # Base class (common functionality)
    │   ├── reviewer.py         # /review - Code review
    │   └── ask.py              # /ask - Q&A
    └── skills/                 # Built-in Skills
        ├── code-review/
        │   ├── SKILL.md        # Review instructions
        │   └── references/     # Reference documents
        └── ask/
            └── SKILL.md

Key Components

Component Purpose Notes
skill_loader.py Manage skills Load SKILL.md, set skills_dir for Agent SDK
base.py Common tool functionality Repo cloning, Agent SDK execution
Agent SDK LLM execution Automatic token management, script execution, context handling, direct Markdown output

FAQ

Q: How to get Kimi API Key?

Visit Moonshot AI Platform, register and create an API Key in the management page. New users get free credits.

Q: Does it support private repositories?

Yes. Just ensure GITHUB_TOKEN has permission to read repository contents.

Q: What if PR is too large?

The Kimi Agent SDK automatically handles large PRs:

  • 256K token context window: Can handle very large PRs
  • Automatic context management: SDK intelligently manages what to include
  • Smart file filtering: Excludes binary files, lock files, minified files

No manual chunking needed - the Agent SDK handles everything automatically.

Q: What is Agent SDK and why use it?

Kimi Agent SDK is an intelligent agent framework that:

  • Automatic token management: No need to manually count tokens or manage context
  • Dynamic script execution: Automatically calls skill scripts when needed
  • Built-in tools: Provides file operations (read/write) and bash execution
  • Context optimization: Intelligently manages conversation context

This allows the action to focus on what to review (skills, rules) rather than how to execute (token counting, script running).

Q: How do skills work with Agent SDK?

Skills define what the agent should do:

  1. SKILL.md contains instructions for the agent
  2. scripts/ contains executable tools (Python scripts)
  3. Agent SDK automatically calls scripts when needed based on instructions

Example flow:

1. Load skill: code-review
2. Pass skills_dir to Agent SDK
3. Agent reads SKILL.md instructions
4. Agent automatically calls scripts/check_security.py when analyzing code
5. Agent generates review based on script output + instructions

Q: How to customize review rules?

Create .kimi-config.yml in your repo root, or add custom Skills in .kimi/skills/ directory. See Configuration section above.

Q: How to use a custom API endpoint?

If you're using a proxy or custom Kimi API endpoint, add KIMI_BASE_URL to your repository secrets:

  1. Go to Settings → Secrets and variables → Actions
  2. Click New repository secret
  3. Add KIMI_BASE_URL with your custom endpoint (e.g., https://your-proxy.example.com/v1)

Then use it in your workflow:

- uses: xiaoju111a/kimi-actions@main
  with:
    kimi_api_key: ${{ secrets.KIMI_API_KEY }}
    kimi_base_url: ${{ secrets.KIMI_BASE_URL }}  # Custom endpoint from secrets
    github_token: ${{ secrets.GITHUB_TOKEN }}

Note: If KIMI_BASE_URL is not set, it defaults to https://api.moonshot.cn/v1.

This is useful for:

  • Using a corporate proxy
  • Testing with a local development server
  • Using alternative API gateways
  • Keeping endpoint URLs private

Acknowledgments

License

MIT

About

AI code review GitHub Action powered by Kimi (Moonshot AI)

Resources

Readme

License

MIT license
Activity
Custom properties

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

9 tags

Packages

 
 
 

Contributors

Languages