AgentSkillsCN

security-review

专门用于安全审查、威胁建模和补救指导的技能。在身份认证/权限变更、密钥或PII处理、公共端点或依赖升级时使用。

SKILL.md
--- frontmatter
name: security-review
description: A specialist skill for security reviews, threat modeling, and remediation guidance. Use for auth/permissions changes, secrets or PII handling, public endpoints, or dependency upgrades.
allowed-tools:
  - Read
  - Grep
  - Glob
  - Bash

Security Review Skill

Tooling Notes

This skill should only use read-only commands and avoid modifying files.

Workflow

Copy this checklist and use it to track your progress through a security review:

markdown
Security Review Checklist

- [ ] Scope the Change
  - [ ] Identify entry points, data flows, and trust boundaries.
  - [ ] Note any changes to auth, secrets, or external integrations.
- [ ] Threat Model
  - [ ] Enumerate likely threats (STRIDE or similar).
  - [ ] Focus on inputs, storage, and outputs.
- [ ] Validate Input & Output
  - [ ] Validate and sanitize user-controlled input.
  - [ ] Encode outputs to prevent injection.
  - [ ] Constrain file paths and URLs (avoid traversal/SSRF).
- [ ] AuthN/AuthZ
  - [ ] Authentication checks are present and consistent.
  - [ ] Authorization rules are explicit and least-privilege.
  - [ ] Session/token handling uses secure defaults.
- [ ] Secrets & PII
  - [ ] No hardcoded secrets or credentials.
  - [ ] Logs avoid sensitive data.
  - [ ] Storage and transport are protected.
- [ ] Dependency Risk
  - [ ] New dependencies are minimal and justified.
  - [ ] Lockfiles updated and reviewed.
- [ ] Verification
  - [ ] Add or update tests for security-sensitive behavior.
  - [ ] Confirm error handling does not leak data.

Security Guardrails

Review Focus Areas

  • •Authentication and authorization checks are explicit and consistent.
  • •Input validation is present on all user-controlled data.
  • •Secrets and credentials never appear in logs or configs.
  • •Public endpoints enforce rate limits and safe defaults.
  • •Dependencies are minimized and vetted for risk.

Common Red Flags

  • •New endpoints without authN/authZ checks.
  • •Raw SQL, command execution, or templating with unsanitized input.
  • •Hardcoded secrets, tokens, or private keys.
  • •Broad permissions or wildcard roles.

Remediation Notes

When issues are found, propose the smallest safe change that addresses the risk, then add or update tests to prevent regressions.