AI资讯大全AIPPXP.CN搜索 ↗
图文教程 · 手把手

Illustrated Tutorial: A Step-by-Step Guide to Conducting Your First Six-Stage Security Audit Using Official Cloudflare Techniques图文教程:手把手教你用 Cloudflare 官方技能跑第一次六阶段安全审计

Follow along step by step: Unzip and install the security-audit skill → Select a small repository for a test run → Understand the division of labour between the reconnaissance ledger and vulnerability hunting → Comprehend the three types of determinations in findings.json and the final report → Master incremental auditing and cost control. Compatible with Claude Code / Codex / Cursor, with zero additional dependencies.

跟着做一遍:解压安装 security-audit 技能 → 选一个小仓库试跑 → 看懂侦察账本与猎洞分工 → 读懂 findings.json 三种判定与最终报告 → 掌握增量审计与成本控制。适配 Claude Code / Codex / Cursor,零额外依赖。

2026-11-13 更新 · 免费
图文教程:手把手教你用 Cloudflare 官方技能跑第一次六阶段安全审计
1

Step 1: Install Skills第一步:安装技能

Download the security-audit skill pack from this site (see the button above) and unzip it. Claude Code users: Copy the entire extracted skills/security-audit directory into ~/.claude/skills/ (for project-level, place it in the project's .claude/skills/). Alternatively, you can use the skills CLI: npx skills add repository address.

After copying, restart the Agent session, enter "Perform a security audit on this repository" or directly specify the skill name to confirm that the Agent can load SKILL.md. The skill includes a zero-dependency Node validation script, so there is no need to install any npm packages.

中文

从本站下载 security-audit 技能包(见上方按钮)并解压。Claude Code 用户:把解压出的 skills/security-audit 整个目录复制进 ~/.claude/skills/(项目级则放进项目 .claude/skills/);也可以用技能 CLI:npx skills add 仓库地址。

复制完重启 Agent 会话,输入「对这个仓库做一次安全审计」或直接点名技能名,确认 Agent 能加载 SKILL.md。技能包含零依赖的 Node 校验脚本,无需安装任何 npm 包。

第一步:安装技能
2

Step 2: Select a small warehouse for a trial run during the reconnaissance phase第二步:选一个小仓库试跑侦察阶段

For the first audit, it is recommended to choose a repository with few dependencies and a small amount of code to control token consumption (a full audit may consume hundreds of thousands of tokens, using your own model quota). The skill will first perform reconnaissance: drawing an architecture diagram, trust boundaries and input surfaces, and generating architecture.md and coverage-ledger.json coverage ledgers.

After the ledger is generated, validate-coverage-ledger.cjs will automatically run to verify its validity. Open these two files and take a look: architecture.md is its understanding of the system, and coverage-ledger.json is the "task list" for subsequent bug hunting. Each unit will be assigned a hunter Agent that is isolated from the others.

中文

第一次审计建议选依赖少、代码量小的仓库,控制 token 消耗(完整审计可能消耗数十万 token,用的是你自己的模型额度)。技能会先做侦察:绘制架构图、信任边界与输入面,生成 architecture.md 和 coverage-ledger.json 覆盖账本。

账本生成后会自动跑 validate-coverage-ledger.cjs 校验合法性。打开这两个文件看一眼:architecture.md 是它对系统的理解,coverage-ledger.json 是后续猎洞的「任务清单」,每个单元都会派一个互相隔离的猎手 Agent。

3

Step 3: Understand the hunting pit and the three judgements第三步:看懂猎洞与三种判定

Hunters search by target type: There are dedicated manuals for 14 attack surfaces, including web protocols and authentication, client-side DOM, supply chain, AI prompt injection, and memory safety. Each candidate vulnerability is handed over to a brand-new validator Agent dedicated to "falsification" to prevent the Agent from generating false positives.

In the final findings.json, there are only three determinations: confirmed (with a complete source code link and bounded empirical results), needs_validation (stating unresolved facts, without severity), and rejected (disproven). After Phase 4 and Phase 5, validate-findings.cjs is run once to enforce validity – machine-readable, reviewable, and not based on verbal claims.

中文

猎手按目标类型翻查:Web 协议与认证、客户端 DOM、供应链、AI 提示注入、内存安全等 14 个攻击面各有专属手册。每个候选漏洞都会交给一个全新的验证者 Agent 专职「证伪」,防止 Agent 自嗨式误报。

最终 findings.json 里只有三种判定:confirmed(有完整源码链路和有界实测结果)、needs_validation(写明未决事实,不带严重度)、rejected(已证伪)。Phase 4 与 Phase 5 之后各跑一次 validate-findings.cjs 强制合法——机器可读、可复核,不靠一张嘴。

4

Step 4: Read the report and perform incremental re-runs第四步:读报告与增量复跑

After running, three reports are generated: REPORT.md contains the conclusions for human review, FINDINGS-DETAIL.md contains all the details, and NEEDS-VALIDATION.md contains items pending manual verification. For each confirmed record, an independent Agent also performs a source verification at the end. For substantial replacements, another verifier will be assigned – every line of code cited in the report is trustworthy.

Running the same repository again is incremental: it uses the previous ledger to target and fill in blind spots, and only re-verifies the source code that has been changed. It is recommended that you commit the generated ledger and findings to the repository, and accumulate them over time to form your project's own security knowledge base.

中文

跑完得到三份报告:REPORT.md 给人看的结论、FINDINGS-DETAIL.md 全部细节、NEEDS-VALIDATION.md 待人工确认项。对每条 confirmed 记录,最后还有独立 Agent 做过来源核验,实质性替换会再派一个验证者——报告里引用的每一行代码都值得信。

同一仓库再次运行是增量的:它用之前的账本定向补盲区、只重验已变更的源码。建议把生成的账本和 findings 提交进仓库,长期积累成你项目自己的安全知识库。

第四步:读报告与增量复跑
5

Problem Encountered遇到问题

Audit interrupted: Mostly due to model quota exhaustion; re-running will automatically continue from the ledger. Skill not loaded: Confirm that the directory name is security-audit and that SKILL.md is located at its root. The skill pack accompanying this article can be found via the download button above.

中文

审计中断:多为模型额度耗尽,重跑会自动接着账本走;技能没被加载:确认目录名是 security-audit 且 SKILL.md 在其根部。本文配套技能包见上方下载按钮。

读完接着看 · Keep reading

想马上用起来?去「AI工具」栏挑一个直接下载,或在「AI教程」里跟着图文步骤做一遍。

去 AI工具 → 看 AI教程 →
0阅读0 条评论

阅读与点赞数据保存在你的浏览器本地,欢迎留下你的想法。

评论 文明发言,让讨论更有价值

正能量公益广告今日正能量学一点,用一点;今天种下的种子,会长成明天的能力。去免费下载专区 →广告