最后更新:11/7/2026最后核验:2026-11-07

Security Audit Skill

Security Audit Skill

0
0

Security Audit Skill 是 Cloudflare 开源的免费 MIT 编码智能体安全审计技能:以侦察、覆盖度引导的猎捕、对抗性候选验证和机器可读 findings.json 输出组织多阶段审计,全部记录由独立的全新智能体二次核验。

免费

<p>Security Audit Skill 把你的编码智能体变成一名纪律严明的安全审计员。审计分六个阶段:先侦察架构、信任边界和输入面,产出 architecture.md 和 coverage-ledger.json;再从台账单元派出相互隔离的猎手,用覆盖度批评者找盲区;每个独特候选交给一个全新的验证者专门试图推翻它;结论按 confirmed、needs_validation、rejected 三档写入 findings.json,并用零依赖校验器对照 report-schema.json 校验;最后由新的独立智能体复核最终来源声明,重大替换还要再审一遍。单次运行大约能找到重复运行总成果的一半,多轮审计是累加的:它会复用此前的台账和结论瞄准缺口、重新验证已变更的源码,且不把过期或未定论的工作当成已覆盖。技能附有内存安全、供应链、云部署、协议、客户端、LLM 注入等攻击类别的狩猎参考文件。使用前提:需要支持工具调用与并行子智能体的编码智能体、Node.js 运行校验器,以及一个禁外网的系统级沙箱来执行目标代码,否则流程会把发现保留为 needs_validation 而不执行目标代码。它按你自己的模型与额度运行,不是一键扫描产品。本文依据 Cloudflare 官方 GitHub 仓库与博客核验,未对任何代码库发起审计。</p><ul><li>六阶段审计流:侦察、覆盖度猎捕、候选验证、结构化输出与独立复核</li><li>confirmed / needs_validation / rejected 三档结论,对照 JSON Schema 自动校验</li><li>多轮审计累加复用台账,瞄准覆盖缺口并重验变更源码</li><li>覆盖内存安全、供应链、云、协议、客户端与 LLM 目标的攻击类参考</li></ul>

详情

平台

macOSWindowsLinux

功能特性

  • Six-phase audit flow: reconnaissance, coverage-led hunting, candidate validation, structured output, and independent record verification
  • confirmed / needs_validation / rejected verdicts validated against report-schema.json with zero-dependency validators
  • Multi-run additive audits reuse prior ledgers and findings to target coverage gaps and revalidate changed source
  • Attack-class reference files for memory safety, supply chain, cloud, protocols, client-side, and LLM-related targets

支持语言

en

适合人群

Security teams wanting reproducible, machine-readable audit workflowsOSS maintainers auditing their own repos before releasesEngineering orgs building internal vulnerability-discovery harnesses

已知限制

  • Needs a coding agent with tool use and parallel sub-agents, plus Node.js for the validators
  • Target-controlled builds and dynamic checks require an OS-enforced sandbox with disabled external networking
  • Single-run audits catch roughly half the vulnerabilities that repeated runs find, per Cloudflare's test runs
  • Runs your agent models under your own plan; not a turnkey scanning product

替代工具

OpenHack — Managed AI security platform with live-app scanning and auto-fix PRs
AiSOC — Self-hosted security-operations center over live security telemetry

为此工具评分

相关工具