AgentSkillsCN

review-security

以安全性为导向的代码审查。检查漏洞、注入攻击、身份验证问题以及数据暴露风险。

SKILL.md
--- frontmatter
name: review-security
description: Security-focused code review. Checks for vulnerabilities, injection attacks, auth issues, and data exposure.

Security Review

Review code changes for security vulnerabilities and unsafe patterns.

Philosophy

Defense in depth. Don't rely on a single security control. Look for places where multiple layers should exist.

Trust boundaries matter. Identify where data crosses trust boundaries (user input, external APIs, database). These are high-risk areas.

Fail secure. When things go wrong, they should fail closed, not open. Check error handling paths.

Checklist

Injection Vulnerabilities

  • • SQL Injection: Are queries parameterized? No string concatenation with user input?
  • • Command Injection: Are shell commands using safe APIs? No exec() with user input?
  • • XSS: Is user content escaped before rendering? Using safe templating?
  • • Template Injection: Are template engines configured safely?
  • • Path Traversal: Are file paths validated? No ../ sequences from user input?

Authentication & Authorization

  • • Auth Checks: Are endpoints properly protected? No missing auth middleware?
  • • Authorization: Are users authorized for the specific resource, not just authenticated?
  • • Session Management: Are sessions handled securely? Proper expiration?
  • • Token Handling: Are tokens stored securely? Not in localStorage for sensitive apps?

Data Exposure

  • • Logging: Are sensitive values (passwords, tokens, PII) excluded from logs?
  • • Error Messages: Do errors expose internal details (stack traces, SQL, paths)?
  • • API Responses: Are responses filtered to exclude sensitive fields?
  • • Source Control: Are secrets kept out of code? Using environment variables?

Cryptography

  • • Hardcoded Secrets: No API keys, passwords, or tokens in code?
  • • Weak Crypto: Using modern algorithms? No MD5/SHA1 for security purposes?
  • • Random Values: Using cryptographic randomness for security-sensitive values?

Dependencies & Configuration

  • • Known Vulnerabilities: Are dependencies up to date? Any known CVEs?
  • • CORS: Is CORS configured restrictively? Not * for sensitive endpoints?
  • • CSRF: Are state-changing requests protected against CSRF?
  • • Headers: Are security headers set (CSP, X-Frame-Options, etc.)?

Severity Guidelines

Blocker (must fix):

  • •Any injection vulnerability
  • •Missing authentication on sensitive endpoints
  • •Exposed secrets or credentials
  • •Direct data exposure of PII

Warning (should fix):

  • •Overly permissive CORS
  • •Missing security headers
  • •Logging sensitive data
  • •Weak but not broken crypto

Note (consider):

  • •Outdated but not vulnerable dependencies
  • •Missing rate limiting
  • •Verbose error messages (non-sensitive)

Output Format

code
## Security Review

### Blockers
- [src/api/users.ts:45] SQL injection: User input concatenated into query
- [src/auth/login.ts:23] Hardcoded API key in source

### Warnings
- [src/server.ts:12] CORS allows all origins (*)
- [src/utils/logger.ts:34] Password field logged in debug mode

### Notes
- Consider adding rate limiting to /api/auth endpoints
- CSP header not set (low risk for API-only backend)

### Verdict: FAIL
Found 2 blockers that must be fixed.