AI代码审查自动化2026：GitHub Actions + GPT-4

在人工审查之前自动审查拉取请求，捕获错误和安全问题。

AI代码审查能发现什么

安全：SQL注入、硬编码密钥、路径遍历

错误：逻辑错误、缺少空值检查、条件错误

性能：N+1查询、不必要的计算

质量：缺少错误处理、代码不清晰

GitHub Actions 工作流

yaml
name: AI Code Review
on:
  pull_request:
    types: [opened, synchronize]
jobs:
  review:
    runs-on: ubuntu-latest
    permissions:
      pull-requests: write
      contents: read
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
        with: {python-version: '3.12'}
      - run: pip install openai PyGithub
      - name: AI Review
        env:
          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
          PR_NUMBER: ${{ github.event.pull_request.number }}
          REPO_NAME: ${{ github.repository }}
        run: python scripts/ai_review.py

审查脚本

python
scripts/ai_review.py
import os
from github import Github
from openai import OpenAI
openai_client = OpenAI(api_key=os.environ['OPENAI_API_KEY'])
gh = Github(os.environ['GITHUB_TOKEN'])
repo = gh.get_repo(os.environ['REPO_NAME'])
pr = repo.get_pull(int(os.environ['PR_NUMBER']))
PROMPT = (
    'You are an expert code reviewer. Analyze this diff for:\n'
    '1. Security: SQL injection, XSS, hardcoded secrets\n'
    '2. Bugs: logic errors, null pointer risks\n'
    '3. Performance: N+1 queries, inefficiencies\n'
    '4. Quality: missing error handling\n'
    'For each issue: file+line, explain problem, suggest fix.\n'
    'Be concise. Focus on significant issues only.\n\nDiff:\n{diff}'
)
def get_diff():
    parts = []
    total = 0
    for f in pr.get_files():
        if not f.patch or f.status == 'removed': continue
        part = f'\n## {f.filename}\n

\n{f.patch[:5000]}\n

'
        total += len(part)
        if total > 80000: break
        parts.append(part)
    return '\n'.join(parts)
def run_review():
    diff = get_diff()
    if not diff.strip(): return
    r = openai_client.chat.completions.create(
        model='gpt-4o',
        messages=[{'role': 'user', 'content': PROMPT.format(diff=diff)}],
        temperature=0.1, max_tokens=2000
    )
    pr.create_issue_comment(
        f'## AI Code Review\n\n{r.choices[0].message.content}\n\n*by GPT-4o*'
    )
    print('Review posted!')
run_review()

安全模式（阻止严重问题）

python
import sys
SECURITY_CHECK = (
    'Scan for security vulnerabilities only.\n'
    'Check: SQL injection, command injection, path traversal, hardcoded secrets, missing auth.\n'
    'Rate each: CRITICAL/HIGH/MEDIUM/LOW\n'
    'If none: respond exactly: No security issues found.'
)
def security_review():
    r = openai_client.chat.completions.create(
        model='gpt-4o',
        messages=[{'role': 'user', 'content': SECURITY_CHECK + '\n\nDiff:\n' + get_diff()}],
        temperature=0
    )
    review = r.choices[0].message.content
    if 'CRITICAL' in review:
        pr.create_review(body=f'## Critical Security Issues!\n\n{review}', event='REQUEST_CHANGES')
        sys.exit(1)  # Fail CI, block merge
    else:
        pr.create_issue_comment(f'## Security Review\n\n{review}')
security_review()

智能模型路由

python
def select_model(pr) -> str:
    critical_patterns = ['auth', 'payment', 'crypto', 'password', 'token', 'secret']
    for f in pr.get_files():
        if any(p in f.filename.lower() for p in critical_patterns):
            return 'gpt-4o'  # Security-critical: use best model
    return 'gpt-4o-mini'  # Routine: use cheaper model

结论

AI代码审查创建了一个强大的自动化质量门。对于常规PR使用gpt-4o-mini，对于安全关键代码使用gpt-4o，并在检测到严重漏洞时阻止合并。