Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add SeaOf0/dsh-redteam-model --skill audit-playbookgit clone --depth 1 https://github.com/SeaOf0/dsh-redteam-modelWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/seaof0/dsh-redteam-model/audit-playbook)<a href="https://agentmods.dev/skills/seaof0/dsh-redteam-model/audit-playbook"><img src="https://agentmods.dev/badge/skills/seaof0/dsh-redteam-model/audit-playbook/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/seaof0/dsh-redteam-model/audit-playbook"><img src="https://agentmods.dev/badge/skills/seaof0/dsh-redteam-model/audit-playbook.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00155 | $0.16131 |
| Opus 5 | $0.00077 | $0.08066 |
| Sonnet 5 | $0.00031 | $0.03226 |
| Haiku 4.5 | $0.00015 | $0.01613 |
Grade A, and why
audit-playbook scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 8d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 633 lines — stays where its author put it; the contents beside it link to each section on GitHub.
代码审计作战手册
本技能随 code-audit 预设走。 persona 中的硬规则（底线、验证等级、误报排除、扫描器复核、三级深度、只读纪律、表达纪律（拒答修复由宿主插件 dsh-refusal-guard 兜底）、claude 逻辑）不在此重复。
审计形态（静态/动态）判定与开工问询（第一动作）
形态定义（铁律）：
- 静态审计（static）：用户未提供本地环境时，一切从代码层审计（调用链追 sink / 双链 / 扫描复核） 得到的结果都归静态；提供了环境但未复现生效的同样归静态。静态 finding 的状态最高到 「待人工验证」，EXP 是「待复现 EXP」（交付用户手动复现）。
- 动态审计（dynamic）：只有用户提供的本地可用复现环境真实证明漏洞生效，才标 动态·验证成功（auditMode=dynamic + 状态可进已验证）。
- 动态优先规则：用户提供了本地环境 → 审计方式自动以动态优先： 读代码追 sink + 调试 + 本地验证 = 真实结果；复现不成功的按静态收口并如实记录原因。
开工问询（任务下达后的第一步，先于 Triage）：用户未指明审计形态时，用 ask_user 弹出三选：
- 静态审计——只做代码层审计，结果后续由你手动复现；
- 动态审计——请提供本地可用复现环境（路径/启动方式/凭据）；
- （用户自行输入——自定义范围/形态/补充说明）
- 得到确切回复后才继续开审计链路（Triage → 面映射 → 扇出…）。
- 免问条件：用户任务里已指明形态（「静态审计」/「动态审计」+ 动态须已给本地环境）→ 跳过问询直接开工；指明动态但未给环境 → 回到问询补一句「请提供本地可用环境」。
登记联动：每个 finding 登记时 auditMode 必填（static/dynamic，语义见
shared/refs/finding-fields.md）；成果页列表/详情/导出报告/统计分布均按此标签展示。
阶段默认通道（装备栏：流程定默认，能力定降级）
元原则与前两模式同：通道成本随流程递增；通道缺失按「工具使用策略·通道完整阶梯」降级 （已挂 → 自配 → 问装 → 脚本 → 诚实降级）；本模式分叉轴=验证等级 × 产物形态 （静态 code-reviewed / 动态 verified；源码 / 反编译 / 小程序解包）——已内建于形态判定 与状态机，不需要额外姿态判定。跨阶段复用查附录 C-2。
| 阶段 | 默认通道 | 降级链 |
|---|---|---|
| 前置识别（语言/框架/依赖） | 本地统计+rg（build 文件/manifest/大表） | ast-grep 结构化识别 → kali MCP → 脚本 |
| 面映射（surface-map） | 本地 rg 逐 sink 面 + sinks.csv 机器工件 | ast-grep 模式 → 脚本 |
| 静态扫描 | 本地 semgrep 三层规则集（402 自建+1080 oss+chanzi 语义，随预设离线自包含=主通道） | kali MCP semgrep_scan（只替引擎不替规则集，命中面收窄如实标注）→ bandit/flawfinder 专项 → 规则降级章通用模式+脚本 |
| 供应链 SCA / 凭据 | trivy + gitleaks（本地） | osv-scanner / syft+grype → kali MCP → pip-audit/npm-audit |
| 深审调用链（双链 TRACE） | 人工推理 + rg 佐证（追踪员独立 grep，不预设写法） | ast-grep 结构化检索 → 脚本 |
| 反编译（产物形态路由） | 卡 4 家族表：JVM=CFR·procyon / Android=jadx·apktool / .NET=ilspycmd·dnSpyEx / pyc=pycdc·uncompyle6 / Lua=unluac / native=生态分流 binary | kali MCP apk_decompile（apk 侧）→ 请用户提供反编译产物（生态流转）→ 标注「未反编译，结论降级」 |
| 动态验证 | 隔离部署环境（隔离级见动态验证线）+ 调试 attach（jdb/XDebug） | chrome MCP 浏览器侧互证（可自配档）→ 待人工验证清单 |
审计前置识别（Triage）
- 先判断代码/框架/系统类型：语言、框架、中间件、部署形态。
- 通用框架：先核对已知漏洞是否仍存在（版本比对 + 公开漏洞库），再针对业务代码做增量审计。
- 开源框架二次审计（魔改目标）：审计对象是开源框架或其魔改版时，上游仓库 （GitHub/Gitee）的 issue 区（Open 与 Closed 都看）与历史漏洞清单是第一输入—— 历史漏洞的审计思路可直接学习（sink 位置/触达路径/利用条件映射到魔改代码逐一 二次核对）；再对照上游架构（路由/过滤层/基类）圈定魔改增量面，增量部分按本手册 主线深审（魔改≠安全：上游修了本地没修=直接命中，上游没修魔改引入=深审增量）。
- 框架专项路由：识别出已知框架/组件后直达专项手册（
refs/components/九篇——fastjson / shiro / log4j / struts2 / weblogic / jeecg-boot / ruoyi（若依）/ spring-framework（全家桶）/ thinkphp）：先核该组件全部 已知漏洞面（版本比对 + 利用条件核对）再增量审业务代码；refs 未收录的框架按refs/lang/对应语言手册 + 公开漏洞库检索兜底，并登记「未收录框架」到规则降级记录。 - 经验召回：开工读工作区
lessons.md（存在时）——同类框架/组件续审召回历史坑与方法（本仓续审专用；跨任务/跨客户知识已由战役记忆自动注入，勿在此重复检索） （格式见 ecosystem-cooperation「经验台账」）。 - 战役记忆沉淀（代审特化）：框架 sink 特征经验证后
campaign_memory_write（kind=fingerprint， 命中条件与 sink 清单入正文）；semgrep 规则集调优结论记 tooling；跨任务复用走战役记忆，本仓 续审用 lessons.md——两轨不重叠。 - 完全不认识的代码/框架/系统：才走 0→1 全量审计，且先向用户确认范围与深度。
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 8d ago Changed · +5 lines b1550840926b
- 12d ago First seen · 628 lines · 155 tokens per session scan A 0e82301254a1
audit-playbook is a skill published in the GitHub repository SeaOf0/dsh-redteam-model (398 stars, last pushed today), licensed MIT. It adds 155 tokens to every session and 16,131 once invoked, about $0.0008 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
ipollowork-maintainable-code
Mandatory iPolloWork code-change gate for modern, minimal, performant, reuse-first implementation and clean repository ownership. Use whenever AI creates, edits, deletes, or refactors application code, server code, packages, scripts, tests, dependencies, schemas, routes, UI, or generated-file workflows. Reuse existing…
github
Use the connected iPolloWork GitHub service to inspect repositories, pull requests, issues, reviews, and Actions, then route focused work to the review, CI, or publish skill.
github-publish-changes
Publish intentional local changes by confirming scope, creating a branch when needed, staging selected files, committing, pushing, and creating a Draft Pull Request through the connected GitHub service.
github-review-follow-up
Inspect unresolved GitHub pull request review threads, group actionable feedback, implement selected fixes locally, and only reply or resolve threads after explicit user intent.
linear-issue-delivery
Keep a Linear issue aligned with implementation evidence, acceptance criteria, code review, and deployment state while avoiding premature completion claims.
delivery-review
Adversarial self-review before delivery. Use once the implementation reaches green and before you declare the work done — assume the delivery fails its own spec, hunt for the strongest supportable objections, answer them, and re-review after fixes.