AgentSkillsCN

post-dev-security-review

在最近的代码变更完成后，开展实施后的安全审查，并结合文件证据与修复建议，优先提出加固方案。在开发任务完成之后、合并/发布之前，或当用户请求安全检查、漏洞分析，或寻求后端/前端/API/Worker代码的安全优化建议时，可使用此功能。

SKILL.md
--- frontmatter
name: post-dev-security-review
description: Perform post-implementation security review on recent code changes and produce prioritized hardening recommendations with file evidence and fix guidance. Use when a development task is completed, before merge/release, or when users ask for security checks, vulnerability analysis, or security optimization suggestions for backend/frontend/API/worker code.

Post Dev Security Review

Overview

Review changed code after implementation and output actionable security improvements, not generic advice. Focus on real attack paths, impact, and concrete mitigation steps tied to specific files.

Review Workflow

  1. •Identify change scope.
  • •Prefer git status --short and git diff --name-only.
  • •If no diff is available, ask the user which files were changed and continue.
  1. •Triage by risk.
  • •Prioritize auth, permission checks, input validation, secrets, file/command execution, DB queries, external HTTP calls, and LLM provider integrations.
  • •Review changed dependencies and lockfiles for supply-chain risk.
  1. •Run a checklist-driven assessment.
  • •Use references/deepseek-oracle-security-checklist.md.
  • •Apply only sections relevant to touched files and stack components.
  1. •Produce a structured report.
  • •Start with findings sorted by severity: Critical, High, Medium, Low.
  • •For each finding include: risk, exploit path, impact, file evidence, and fix guidance.
  • •If no blocking issue is found, still provide hardening improvements and tests.

Output Contract

Use this structure exactly:

Findings

  • •[Severity] Title
  • •Risk: What can go wrong.
  • •Attack path: How an attacker could exploit it.
  • •Impact: Data, integrity, availability, or cost consequences.
  • •Evidence: path/to/file:line.
  • •Fix: Specific remediation; include patch-style suggestions when useful.

Security Optimization Suggestions

  • •Provide 3-5 concrete improvements, prioritized by risk reduction and implementation cost.
  • •Tie each suggestion to one or more changed files.

Verification Steps

  • •List commands/tests to validate the mitigation.
  • •Include at least one negative test (malicious input or unauthorized case).

Quality Rules

  • •Avoid vague statements like "improve security" without code-level direction.
  • •Do not report speculative vulnerabilities without evidence; mark assumptions explicitly.
  • •Prefer smallest safe change first, then optional stronger hardening.
  • •Always mention secret handling and sensitive logging exposure if relevant to changed files.
  • •For LLM/provider code, check key leakage, prompt injection exposure, unsafe tool access, and unbounded external calls.

References

  • •Read references/deepseek-oracle-security-checklist.md before finalizing conclusions.