Code Review

Code Review Skill

Complete 5-step code review workflow for GitCode PRs.

5-Step Review Process

Step 1: Automated Scanning

Run script to detect critical issues:

python scripts/review_pr.py <pr_url> [token]

Detects: SQL injection, command injection, XSS, eval(), hardcoded credentials, resource leaks, infinite loops.

Output: review_result.json

Step 2: Manual Review (REQUIRED)

Always read all changed code manually. Script misses:

• Logic errors and edge cases
• Design flaws
• Performance issues
• Missing error handling
• Business logic errors
• Code duplication
• Test coverage gaps

How to get diff:

curl -H "Authorization: Bearer <token>" \
  "https://gitcode.com/api/v5/repos/<owner>/<repo>/pulls/<number>/diff"

Important: For each issue found, record:

• File path: e.g., src/components/Table.tsx
• Line range: e.g., L42-L45 (the line numbers of the problematic code)
• Problem code: The actual code snippet
• Description: Detailed explanation of the issue
• Suggestion: Specific fix recommendation

Step 3: Select Top 3 Issues

Combine automated + manual findings:

• Filter false positives from script
• Add issues found in manual review
• Sort by severity (1-10)
• Select top 3 most important

Generate json format file top3_issues.json for these 3 issues to use in next step.

top3_issues.json must be created in the directory of format_review.py for the next step to read.

Important:

• The description field must contain the complete description from Step 1 and Step 2 findings, not a simplified version. Include all context and details.
• The position field must be the last line number of the problematic code range (e.g., if problem code is at L42-L45, use 45)

Structure:

{
  "meta": {
    "total_issues": 5,
    "selected_issues": 3,
    "automated_count": 2,
    "manual_count": 3
  },
  "top3_issues": [
    {
      "number": 1,
      "path": "src/file.py",
      "position": 45,
      "severity": 8,
      "type": "安全问题",
      "description": "Complete description from Step 1/2 findings, not simplified",
      "suggestion": "Detailed suggestion with specific actions",
      "code": "problematic code snippet from L42-L45",
      "code_context": ""
    }
  ]
}

Note: position uses the last line of the code range for GitCode API positioning.

After generating top3_issues.json, display the top 3 issues in Markdown format:

Top 3 Issues Selected

🔴 问题 #1 | 可维护性问题 | 6/10

文件: server/src/.../CheckProjectValidHandler.cpp**

问题代码行: L119-L124

问题代码:

bool CheckProjectValidHandler::CheckPathSafety(
    const std::string& path,
    ProjectErrorType& error)
{
    ...
}

🟠 问题 #2 | 测试覆盖问题 | 6/10

文件: server/src/.../CheckProjectValidHandler.cpp**

问题代码行: L119

问题代码:

bool CheckProjectValidHandler::CheckPathSafety

🟡 问题 #3 | 代码一致性问题 | 5/10

文件: server/src/.../TimelineProtocolRequest.h**

问题代码行: L68-L72

问题代码:

bool isSafePath = std::any_of(path.begin(), path.end(), ...)

Total: 3 issues selected

Note: position in JSON uses the last line number (e.g., L119-L124 → position: 124)

Step 4: Format Output

Format issues to structured JSON:

python scripts/format_review.py <top3_issues.json> [output.json]

Input:

• top3_issues.json from Step 3

Output: formatted_review.json

formatted_review.json must be created in the directory of post_review.py for the next step to read.

Structure:

{
  "comments": [
    {
      "number": 1,
      "path": "src/file.py",
      "position": 42,
      "severity": 8,
      "type": "安全问题",
      "body": "【review】..."
    }
  ]
}

Comment Format (in body field):

【review】{问题类型}。{问题描述}。{修改建议}。

After generating formatted_review.json, display the formatted content:

Step 4: Formatted Review Comments (Ready to Post)

以下 3 条评论将提交到 PR:

1. `CheckProjectValidHandler.cpp:119`
   类型: 可维护性问题 | 严重程度: 6/10
   内容: 【review】代码重复，违反DRY原则...

2. `CheckProjectValidHandler.cpp:119`
   类型: 测试覆盖问题 | 严重程度: 6/10
   内容: 【review】缺少单元测试...

3. `TimelineProtocolRequest.h:68`
   类型: 代码一致性问题 | 严重程度: 5/10
   内容: 【review】逻辑不一致...

Output: formatted_review.json

Step 5: Post to PR (Optional)

Preview and confirm before posting:

python scripts/post_review.py <owner> <repo> <pr_number> <token> [formatted_review.json]

Parameters:

• owner: Repository owner (e.g., Ascend)
• repo: Repository name (e.g., msinsight)
• pr_number: PR number (e.g., 277)
• token: GitCode access token
• formatted_review.json: Output from Step 4 (default: formatted_review.json)

Example:

python scripts/post_review.py Ascend msinsight 277 your_token_here formatted_review.json

Flow:

1. Read formatted_review.json from Step 4
2. Display preview of all comments
3. Wait for user confirmation (yes/no)
4. Only post if user confirms

Note: Only posts individual issue comments, no summary comment.

API Reference: If unsure how to post PR comments, read API.md for detailed API documentation.

Severity Scale

Manual Review Checklist

Logic & Correctness

• Edge cases (null, empty, max values)
• Error handling paths
• Concurrency/thread safety
• Resource cleanup

Design & Architecture

• Single responsibility
• No code duplication
• Clean interfaces
• Clear dependencies

Performance

• Algorithm complexity
• N+1 queries
• Large data handling
• Memory usage

Security

• Input validation
• Output encoding
• Authorization checks
• Sensitive data handling

Testing

• Tests cover changes
• Edge cases tested
• Error paths tested

API Reference

• Get PR files: GET /api/v5/repos/{owner}/{repo}/pulls/{number}/files
• Get diff: GET /api/v5/repos/{owner}/{repo}/pulls/{number}/diff
• Post comment: POST /api/v5/repos/{owner}/{repo}/pulls/{number}/comments

Scripts