AgentSkillsCN

security-review

对代码库或功能进行系统性的安全审计

SKILL.md
--- frontmatter
name: security-review
description: Systematic security audit of codebase or feature
allowed-tools: Read, Grep, Glob, Bash

Perform a security review of the codebase or a specific feature.

Instructions

  1. •Read .claude/rules/agent-guide.md for project context and tech stack.
  2. •Read docs/spec/.llm/PROGRESS.md for known security considerations.
  3. •Perform a systematic security assessment covering the categories below.

Input Validation & Injection

  • • All user input is validated at system boundaries (handlers, CLI, config)
  • • SQL queries use parameterized statements (no string concatenation)
  • • Command execution inputs are properly escaped
  • • File paths from user input are sanitized (no path traversal)
  • • HTML output is escaped to prevent XSS
  • • JSON/XML parsing has size limits and depth limits

Authentication & Authorization

  • • Auth checks exist on all protected endpoints
  • • Token/session handling follows best practices (expiry, rotation, secure storage)
  • • Password handling uses proper hashing (bcrypt/argon2/scrypt, not MD5/SHA)
  • • Rate limiting exists on auth endpoints
  • • Privilege escalation paths are checked
  • • CORS policy is properly configured

Data Protection

  • • Sensitive data is not logged (passwords, tokens, PII)
  • • Secrets are not hardcoded in source (check for API keys, passwords, tokens)
  • • Environment variables used for secrets, not config files committed to git
  • • Data at rest encryption where required
  • • Data in transit uses TLS
  • • PII handling complies with requirements (GDPR, CCPA, etc.)

Dependencies

  • • No known vulnerable dependencies (run go vuln check, npm audit, etc.)
  • • Dependencies are pinned to specific versions
  • • Minimal dependency surface (no unnecessary packages)
  • • No dependencies with concerning permissions or behaviors

Error Handling & Information Disclosure

  • • Error messages don't leak internal details to external users
  • • Stack traces are not exposed in production responses
  • • Debug endpoints are disabled in production
  • • Logging doesn't include sensitive request/response bodies

Infrastructure

  • • Docker images use non-root users
  • • Container images are minimal (no unnecessary tools)
  • • Network policies limit container communication
  • • Health check endpoints don't expose sensitive info
  • • Resource limits are set (CPU, memory)

Report Format

Present findings as:

code
## Security Review: {scope}

### Summary
{overall assessment: clean / issues found / critical issues}

### Findings
| # | Finding | Severity | Category | Location |
|---|---------|----------|----------|----------|
| 1 | {finding} | {critical/high/medium/low/info} | {category} | {file:line} |

### Detailed Findings

#### Finding N: {title}
- **Severity**: {critical/high/medium/low/info}
- **Category**: {input-validation/auth/data-protection/deps/disclosure/infra}
- **Location**: `{file:line}`
- **Description**: {what's wrong}
- **Impact**: {what could happen if exploited}
- **Recommendation**: {how to fix}
- **Example fix**: {code snippet if applicable}

### Recommendations (Prioritized)
1. **[Critical]** {fix immediately}
2. **[High]** {fix before next release}
3. **[Medium]** {fix soon}
4. **[Low]** {fix when convenient}

$ARGUMENTS