Kimi Code Reviewer

AI-powered code review for GitHub using Moonshot Kimi with 256K context window
基於 Kimi 大模型的 GitHub 智能代碼審查工具，支援 256K 超長上下文

GitHub Action · GitHub App · Configuration · 中文說明

Note: This is a community-maintained GitHub Action for PR code review. For the official Kimi Code CLI with full features (chat, flows, MCP, etc.), see MoonshotAI/kimi-cli.

Features

256K Context Window — Reviews entire PRs with full file context, not just diffs
GitHub Action + App — Use as a CI/CD Action or a self-hosted GitHub App (@kimi mentions)
Prefix Cache — Server-side caching at $0.10/M tokens (75% cheaper than standard $0.39/M)
Inline Annotations — Issues appear directly in the PR diff via GitHub Checks API
Per-Repo Config — .kimi-review.yml for custom rules, severity thresholds, file filters
Multilingual — Review comments in English, 繁體中文, 简体中文, 日本語, 한국어

Quick Start — GitHub Action

1. Get an API Key

Get your API key from Kimi Code Console (recommended for code review use cases).

Kimi Code API vs Moonshot API: The Kimi Code API (api.kimi.com/coding/v1) is optimized for coding tasks and is separate from the general Moonshot API (api.moonshot.cn/v1). Both work with this action, but the Kimi Code API provides better code-specific performance.

2. Add the Secret

Go to your repo Settings → Secrets and variables → Actions and add:

Secret	Value
`MOONSHOT_API_KEY`	Your `sk-...` API key

3. Create the Workflow

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

on:
  pull_request:
    types: [opened, synchronize]

permissions:
  contents: read
  pull-requests: write
  checks: write

jobs:
  review:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: howardpen9/kimi-code-reviewer@main
        with:
          kimi_api_key: ${{ secrets.MOONSHOT_API_KEY }}

That's it. Every PR will now get an AI code review.

Action Inputs

Input	Required	Default	Description
`kimi_api_key`	Yes	—	Moonshot AI API key or Kimi Code API key
`github_token`	No	`${{ github.token }}`	GitHub token for API access
`base_url`	No	`https://api.moonshot.cn/v1`	API endpoint. Set to `https://api.kimi.com/coding/v1` to use Kimi Code
`model`	No	`kimi-k2.5` (or `kimi-for-coding` when `base_url` points at Kimi Code)	Model ID. Auto-defaults based on `base_url`
`language`	No	`en`	Review language: `en`, `zh-TW`, `zh-CN`, `ja`, `ko`
`fail_on`	No	`critical`	Fail the check on: `critical`, `warning`, `never`
`config_path`	No	`.kimi-review.yml`	Path to config file

Example: Use Kimi Code API (recommended for code review)

- uses: howardpen9/kimi-code-reviewer@v2
  with:
    kimi_api_key: ${{ secrets.KIMICODE_API_KEY }}
    base_url: https://api.kimi.com/coding/v1
    # model defaults to kimi-for-coding when base_url is api.kimi.com/coding

Get the Kimi Code API key at code.kimi.com → Settings → API Keys. Kimi Code API keys and Moonshot API keys are not interchangeable.

Action Outputs

Output	Description
`review_summary`	Review summary text
`annotations_count`	Number of inline annotations created
`critical_count`	Number of critical issues found
`tokens_used`	Total tokens consumed
`cost_estimate`	Estimated API cost (USD)

Example: Fail on Warnings + Chinese Reviews

- uses: howardpen9/kimi-code-reviewer@main
  with:
    kimi_api_key: ${{ secrets.MOONSHOT_API_KEY }}
    language: zh-TW
    fail_on: warning

Self-Hosted GitHub App

For teams that want @kimi review commands in PR comments.

Setup

git clone https://github.com/howardpen9/kimi-code-reviewer.git
cd kimi-code-reviewer
npm install
cp .env.example .env  # Fill in credentials
npm run dev

Environment Variables

Variable	Required	Description
`KIMI_API_KEY`	Yes	Moonshot AI API key
`GITHUB_APP_ID`	Yes	Your GitHub App ID
`GITHUB_PRIVATE_KEY`	Yes	GitHub App private key (PEM)
`GITHUB_WEBHOOK_SECRET`	Yes	Webhook verification secret
`PORT`	No	Server port (default: `3000`)
`LOG_LEVEL`	No	Log level (default: `info`)

@kimi Commands

Comment on any PR to trigger:

Command	Description
`@kimi review`	Run a full code review
`@kimi help`	Show available commands

Webhook Events

Event	Trigger
`pull_request.opened`	PR created
`pull_request.synchronize`	New commits pushed
`pull_request.review_requested`	Review requested
`issue_comment.created`	`@kimi` mentioned in comment

Configuration

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

# Review language (en, zh-TW, zh-CN, ja, ko)
language: zh-TW

# Kimi model
model: kimi-k2.5

review:
  # Auto-trigger settings
  auto:
    enabled: true
    onOpen: true        # Review when PR is opened
    onPush: true        # Review on new commits
    onReviewRequest: true
    drafts: false       # Skip draft PRs

  # What to check
  aspects:
    bugs: true
    security: true
    performance: true
    style: true
    bestPractices: true
    documentation: false
    testing: false

  # Severity filter: critical | warning | suggestion | nitpick
  minSeverity: suggestion

  # Max annotations per review (1-100)
  maxAnnotations: 30

  # When to fail the check: critical | warning | never
  failOn: critical

# File filters (minimatch glob patterns)
files:
  include:
    - "**/*"
  exclude:
    - "**/node_modules/**"
    - "**/dist/**"
    - "**/*.lock"
    - "**/*.min.*"
  maxFileSize: 100000  # bytes

# Custom review rules
rules:
  - name: no-console-log
    description: "No console.log in production code"
    severity: warning
    filePattern: "src/**/*.ts"

  - name: input-validation
    description: "All API endpoints must validate input"
    severity: critical
    filePattern: "src/routes/**"

# Custom prompt additions
prompt:
  systemAppend: "Pay special attention to SQL injection risks"
  reviewFocus: "Focus on API input validation and error handling"

# Prefix cache settings
cache:
  enabled: true
  ttl: 3600

If no config file is found, sensible defaults are used.

How It Works

PR Event → Extract Context → Pack (256K) → Kimi API → Parse → Annotations

Review Pipeline (12 Steps)

Create GitHub Check Run (in-progress)
Extract PR metadata, diff, and changed files
Filter files by include/exclude patterns
Pack context with 256K optimization
Build cache-optimized message order
Call Kimi API with structured JSON output
Parse response (supports multiple output formats)
Filter annotations by minimum severity
Limit to max annotation count
Determine pass/fail conclusion
Update Check Run with inline annotations
Post PR review comment with summary

Context Packing Strategies

Automatically selects the best strategy based on PR size:

PR Size	Strategy	What Gets Sent
Small (<50K tokens)	Full	All file contents + full diff
Medium (50–150K)	Mixed	Most-changed files in full, rest as diff only
Large (>150K)	Chunked	Diff only, no file contents

Cost

Kimi's server-side prefix caching reduces repeat costs by ~75%:

Scenario	Estimated Cost
First review of a PR	~$0.01–0.02
Subsequent pushes (cache hit)	~$0.003–0.01

Token Type	Rate
Input	$0.39 / 1M tokens
Output	$1.90 / 1M tokens
Cached input	$0.10 / 1M tokens

Architecture

kimi-code-reviewer/
├── action/              # GitHub Action entry point
│   ├── action.yml       # Action metadata (inputs/outputs)
│   ├── index.ts         # Action runner
│   └── dist/            # Bundled Action (ncc)
├── src/
│   ├── index.ts         # Hono server entry point
│   ├── app.ts           # GitHub App initialization
│   ├── kimi/
│   │   ├── client.ts          # Moonshot API client
│   │   ├── context-packer.ts  # 256K context optimization
│   │   ├── cache-strategy.ts  # Prefix cache message ordering
│   │   ├── prompt-builder.ts  # System/user prompt construction
│   │   └── response-parser.ts # Robust JSON extraction
│   ├── github/
│   │   ├── webhooks.ts   # Event handlers
│   │   ├── pulls.ts      # PR data extraction
│   │   ├── checks.ts     # Check Run API (annotations)
│   │   └── comments.ts   # PR review comments
│   ├── review/
│   │   ├── orchestrator.ts    # 12-step review pipeline
│   │   ├── file-filter.ts     # Glob-based file filtering
│   │   ├── diff-analyzer.ts   # Unified diff parsing
│   │   └── summary-builder.ts # Markdown summary generation
│   ├── config/
│   │   ├── schema.ts     # Zod config validation
│   │   ├── defaults.ts   # Default config values
│   │   └── loader.ts     # YAML config loader from repo
│   └── utils/
│       ├── tokens.ts     # Token estimation + cost calculation
│       ├── logger.ts     # Pino structured logging
│       └── errors.ts     # Custom error classes
├── test/
│   └── unit/             # Vitest unit tests
├── .kimi-review.yml      # Self-review config
└── .github/workflows/
    ├── ci.yml            # CI: lint + test + build
    └── kimi-review.yml   # Self-review on PRs

Development

npm run dev          # Start dev server (hot reload)
npm test             # Run tests
npm run build        # Compile TypeScript
npm run build:action # Bundle GitHub Action with ncc
npm run lint         # Type check only

Severity Levels

Level	Meaning	Example
`critical`	Must fix before merge	Bugs, security vulnerabilities, data loss
`warning`	Should fix	Performance issues, bad practices
`suggestion`	Nice to have	Readability, maintainability improvements
`nitpick`	Optional	Style preferences, minor formatting

Score

Each review includes a quality score (0–100):

Range	Rating
90–100	Excellent
70–89	Good
50–69	Needs improvement
< 50	Significant issues

中文說明

簡介

Kimi Code Reviewer 是基於 Moonshot Kimi 大模型的 GitHub 代碼審查工具。與其他 AI code review 工具不同，Kimi 擁有 256K token 的超長上下文窗口，可以讀取 PR 中所有檔案的完整內容（不僅僅是 diff），提供更準確、更有上下文的審查建議。

特色

超長上下文：256K token 窗口，可以同時理解多個檔案之間的關聯
多語言支援：審查意見支援繁體中文、簡體中文、英文、日文、韓文
行內標註：問題直接標註在 PR diff 的對應行上
智能快取：Prefix caching 讓重複審查節省 75% 費用
靈活配置：每個 repo 可以用 .kimi-review.yml 自訂審查規則

快速開始

1. 取得 API Key

前往 Kimi Code Console 取得 API key（推薦，專為程式碼任務優化）。也可使用 platform.moonshot.ai 的通用 API key。

2. 設定 GitHub Secret

在你的 repo 中前往 Settings → Secrets and variables → Actions，新增：

MOONSHOT_API_KEY：你的 Moonshot API key（sk-... 開頭）

3. 建立 Workflow

在你的 repo 中建立 .github/workflows/kimi-review.yml：

name: Kimi Code Review

on:
  pull_request:
    types: [opened, synchronize]

permissions:
  contents: read
  pull-requests: write
  checks: write

jobs:
  review:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: howardpen9/kimi-code-reviewer@main
        with:
          kimi_api_key: ${{ secrets.MOONSHOT_API_KEY }}
          language: zh-TW    # 使用繁體中文
          fail_on: never     # 不讓 check 失敗

4. 自訂配置（選填）

在你的 repo 根目錄建立 .kimi-review.yml：

language: zh-TW
model: kimi-k2.5

review:
  aspects:
    bugs: true           # 檢查 bug
    security: true       # 檢查安全漏洞
    performance: true    # 檢查效能問題
    style: true          # 檢查程式風格
    bestPractices: true  # 檢查最佳實踐
  minSeverity: suggestion  # 最低回報等級
  maxAnnotations: 30       # 最多標註數量
  failOn: critical         # 只有 critical 時才 fail

files:
  exclude:
    - "**/dist/**"
    - "**/*.lock"

# 自訂規則
rules:
  - name: 禁止 console.log
    description: "正式程式碼不應包含 console.log"
    severity: warning
    filePattern: "src/**/*.ts"

費用估算

場景	預估費用
首次 PR 審查	~$0.01–0.02
同一 PR 後續 push（快取命中）	~$0.003–0.01

嚴重程度說明

等級	說明	範例
`critical`	必須在合併前修復	Bug、安全漏洞、資料遺失風險
`warning`	應該修復	效能問題、不良實踐
`suggestion`	建議改善	可讀性、可維護性提升
`nitpick`	可選	風格偏好、格式微調

License

MIT

Name		Name	Last commit message	Last commit date
Latest commit History 15 Commits
.github/workflows		.github/workflows
action		action
src		src
test/unit		test/unit
.env.example		.env.example
.gitignore		.gitignore
.kimi-review.yml		.kimi-review.yml
LICENSE		LICENSE
README.md		README.md
action.yml		action.yml
package-lock.json		package-lock.json
package.json		package.json
tsconfig.json		tsconfig.json
vitest.config.ts		vitest.config.ts

Folders and files

Latest commit

History

Repository files navigation

Kimi Code Reviewer

Features

Quick Start — GitHub Action

1. Get an API Key

2. Add the Secret

3. Create the Workflow

Action Inputs

Example: Use Kimi Code API (recommended for code review)

Action Outputs

Example: Fail on Warnings + Chinese Reviews

Self-Hosted GitHub App

Setup

Environment Variables

@kimi Commands

Webhook Events

Configuration

How It Works

Review Pipeline (12 Steps)

Context Packing Strategies

Cost

Architecture

Development

Severity Levels

Score

中文說明

簡介

特色

快速開始

1. 取得 API Key

2. 設定 GitHub Secret

3. 建立 Workflow

4. 自訂配置（選填）

費用估算

嚴重程度說明

License

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases 1

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages