ai-review-pipeline

AI-powered code quality pipeline CLI — Review + Test + Report in one command. Add --fix for auto-fix loop.

English | 中文

Features

• Zero config — Built-in free AI model, npx ai-review-pipeline works out of the box with no API Key
• Streaming output — Real-time token-by-token response, no more staring at a blank screen
• Large diff chunking — Auto-splits oversized diffs by file for complete coverage
• Zero dependencies — No required deps, instant npx execution
• Multi-provider — OpenAI / DeepSeek / Claude / Qwen / Gemini / SiliconFlow / Ollama, auto-detected
• Unified pipeline — Default: Review + Test + Report (read-only). --fix enables auto-fix loop
• Flexible targets — Files, folders, comma-separated multi-targets
• --full mode — Review entire file content without git changes
• HTML reports — Score + issue list + fix suggestions, attachable to PRs
• JSONC config — .ai-pipeline.json with comments support, shared across team
• CI-ready — --json output + exit codes for GitHub Actions / GitLab CI
• Bilingual prompts — --lang en switches both output AND AI prompt to English
• Multi-language code — TypeScript / JavaScript / Vue / Python / Go / Rust / Java / Swift / PHP / Kotlin

Quick Start

# Zero config — just run it (built-in free model)
npx ai-review-pipeline

# Auto-fix pipeline
npx ai-review-pipeline --fix

💡 No API Key needed to try the full pipeline. The built-in free model has rate limits — configure your own Key for a faster and more stable experience.

# Pick any provider and add to .env.local

# DeepSeek (affordable & capable)
echo 'DEEPSEEK_API_KEY=sk-xxx' >> .env.local

# OpenAI
echo 'OPENAI_API_KEY=sk-xxx' >> .env.local

# Claude
echo 'ANTHROPIC_API_KEY=sk-ant-xxx' >> .env.local

# Qwen (Alibaba Cloud)
echo 'DASHSCOPE_API_KEY=sk-xxx' >> .env.local

# Google Gemini
echo 'GEMINI_API_KEY=xxx' >> .env.local

# Local Ollama (no key needed)
echo 'AI_REVIEW_PROVIDER=ollama' >> .env.local

Install

# Project-level (recommended for teams)
npm install -D ai-review-pipeline

# Global
npm install -g ai-review-pipeline

# No install needed
npx ai-review-pipeline

Short alias: ai-rp can be used instead of ai-review-pipeline.

Pipeline Design

Default Mode (Review + Test + Report)

① Deterministic Rule Check + AI Review (1 round, read-only)
       │
② AI Test Case Generation (functional / adversarial / edge)
       │
③ Generate HTML Report
       │
④ Gate Check (any 🔴 / 🟡 > maxMajor / score < threshold → exit 1; else exit 0)

--fix Mode (Review + Fix Loop + Test + Report)

① AI Review (score + issue list)
       │
       ├─ Pass ──→ ④ Test
       │
       └─ Fail → ② AI Auto-fix (shows git diff)
                        │
                        └→ ③ Re-review (up to N rounds)
                                │
                                ├─ Pass → ④
                                └─ maxRounds reached → ④ (still generates test + report)

④ AI Test Case Generation
       │
⑤ Generate HTML Report
       │
⑥ Auto git commit (only on pass)
       │
⑦ Exit (pass → exit 0; fail → exit 1)

Full Command Reference

Commands

Core Options

Review Options

Gate Options (shared by review and fix)

Fix Options

Exit Codes

--file vs --full

# Review only git changes for that file
ai-rp --file src/utils.ts

# Review entire file content (no git changes needed)
ai-rp --file src/utils.ts --full

# No --file: review all staged / HEAD git changes
ai-rp

test — AI Test Case Generation

Generate three types of test cases for target files.

ai-rp test --file src/utils.ts
ai-rp test --staged

Three Test Types

init — Initialize Config

ai-rp init    # Creates .ai-pipeline.json in project root

Supported AI Providers

Auto-detection

The tool auto-detects your provider:

• Key starting with sk-ant- → Claude
• DEEPSEEK_API_KEY set → DeepSeek
• ANTHROPIC_API_KEY set → Claude
• DASHSCOPE_API_KEY set → Qwen
• GEMINI_API_KEY set → Gemini

Manual override:

# .env.local
AI_REVIEW_PROVIDER=deepseek
DEEPSEEK_API_KEY=sk-xxx

# Or custom OpenAI-compatible API
AI_REVIEW_PROVIDER=custom
AI_REVIEW_API_KEY=sk-xxx
AI_REVIEW_BASE_URL=https://your-api.com/v1
AI_REVIEW_MODEL=your-model

Ollama (Local)

# 1. Install Ollama: https://ollama.com
# 2. Pull a model
ollama pull qwen2.5-coder

# 3. Configure
echo 'AI_REVIEW_PROVIDER=ollama' >> .env.local

# 4. Run
npx ai-review-pipeline --file src/utils.ts --full

Configuration

Run ai-rp init to generate .ai-pipeline.json. Commit it to share with your team:

{
  "review": {
    "threshold": 85,           // Score threshold (0-100)
    "maxMajor": 3,             // Max 🟡 Majors before blocking merge
    "maxRounds": 5,            // Max fix rounds in --fix mode
    "model": "",               // Override model (default: gpt-4o-mini)
    "maxDiffLines": 1500,      // Auto-truncate diff beyond this line count
    "customRules": [           // Project-specific review rules
      "No any type allowed",
      "API keys must not be hardcoded"
    ]
  },
  "fix": {
    "safetyMinRatio": 0.5      // Fixed file must be at least 50% of original (prevents truncation)
  },
  "test": {
    "stack": "auto",           // Auto-detect tech stack, or specify manually
    "maxCases": 8              // Max test cases to generate
  },
  "report": {
    "outputDir": ".ai-reports",
    "open": true               // Auto-open report after generation
  }
}

Severity & Scoring (Commercial Gate)

Base score: 100. Default pass threshold: 85.

Gate Logic (three-way check)

1. Any 🔴 Blocker → BLOCKED (reason shown: blocker:1(security:1))
2. 🟡 Major > --max-major (default 3) → BLOCKED
3. Score < --threshold (default 85) → BLOCKED
4. All pass → PASS

Issue Categories

Each issue is auto-classified into one of: security | correctness | stability | logic | perf | types | api | ux | quality | style. Categories are shown in the JSON output and HTML report for team triage.

Environment Variables

Supports .env.local and .env auto-loading. Just set the Key for your provider — the tool handles the rest.

CI/CD Integration

GitHub Actions

- name: AI Code Review
  run: npx ai-review-pipeline --json
  env:
    OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}

GitLab CI

ai-review:
  script:
    - npx ai-review-pipeline --json
  variables:
    OPENAI_API_KEY: $OPENAI_API_KEY
  allow_failure: false

Git Hook (lefthook)

# lefthook.yml
pre-push:
  commands:
    ai-review:
      run: npx ai-rp --fix --max-rounds 3

package.json Scripts

{
  "scripts": {
    "review": "ai-rp",
    "review:full": "ai-rp --file src/ --full",
    "review:fix": "ai-rp --fix",
    "review:fix:full": "ai-rp --fix --file src/ --full",
    "test:ai": "ai-rp test --staged"
  }
}

Cheat Sheet

# ── Default mode (Review + Test + Report, read-only) ──
ai-rp                                          # review git changes
ai-rp --file src/a.vue                         # target file
ai-rp --file src/a.vue --full                  # review full file
ai-rp --file src/views --full                  # review entire folder
ai-rp --branch main                            # compare branch
ai-rp --staged --json                          # CI mode

# ── Fix mode (Review + Auto-fix + Test + Report) ──
ai-rp --fix                                    # full fix pipeline
ai-rp fix                                      # alias
ai-rp --fix --file src/a.vue --full            # fix specific file
ai-rp fix --threshold 90 --max-rounds 3        # custom params
ai-rp fix --max-major 5                        # raise major issue tolerance
ai-rp fix --no-commit --skip green             # no commit, fix red+yellow only

# ── Standalone test ──
ai-rp test --file src/utils.ts                 # generate tests
ai-rp test --staged                            # staged file tests

# ── Init ──
ai-rp init                                     # generate config file

Community

• Linux DO Discussion — Chinese developer community

License

MIT