Package all 85 security research skills from upstream reverse-skill for DeepSeek Harness, one-click enable.
- Language
- PowerShell
- License
- MIT
- Branch
- main
Install
$ dsh plugin --profile web add @dhicoc/dsh-reverse-skillRun the command above in your terminal to install this plugin via the dsh CLI. You can switch Profile in the top-right corner. New to dsh? Read the beginner tutorial
Install via your agent
Install the DeepSeek Harness plugin dhicoc/dsh-reverse-skill for me: review the repository at https://github.com/dhicoc/dsh-reverse-skill first, then run the install command and verify the plugin loads successfully.
Paste this instruction to the DSH Web GUI assistant — it will install and verify for you.
One-Line Pitch
Packages the 85 security research skills (43 domain skills + 42 CTF track skills) from the upstream reverse-skill project as-is into a dsh Cordis plugin. After installation, skills are automatically injected into the skill library when the host starts, eliminating the hassle of manually maintaining a candidate list.
Core Features
- One-click mounting of 85 reverse engineering/penetration testing/forensics/firmware/mobile/CTF security research skills into the dsh skill library
- Recursively scans all SKILL.md files under
skills/andCTF-Sandbox-Orchestrator/directories — adding or removing skills requires no plugin code changes - Parses SKILL.md frontmatter (BOM/CRLF compatible), promoting
metadata.user-invocableto top-leveluser-invocable - Registers a skill provider under the
reverse-skillnamespace, callable by both model and user by name - Normalizes upstream camelCase fields like
when_to_useto dsh-supportedwhenToUse, preventing dsh from rejecting the load - Provides BOM/CRLF fixture self-test script to ensure the scanner doesn't silently skip files with BOM
Technical Implementation
- Language: TypeScript (ES2022 / NodeNext, compiled to
lib/index.js+lib/types/index.d.ts) - Key Dependencies:
@deepseek-ai/cordis,@deepseek-ai/dsh-skill(peerDependencies, provided by dsh at runtime) - Architecture Pattern: Cordis plugin —
apply(ctx)callsctx.skills.registerProvider()to register a data-drivenSkillProviderinto the host; scanner usesnode:fs/promises+node:url+node:pathto recursively traverse package directories, results cached in module-levelCACHE - Entry Point:
src/index.ts(exportsname = 'reverse-skill',inject = ['skills'],apply(ctx)); at runtime, loaded by dsh via the bundle entry declared incordis.patch.yml - Auxiliary Files:
port.pynormalizes frontmatter and restructures routing skills into subdirectories when搬运ing the upstream reverse-skill repo to this repo;_selftest.mjstemporarily writes SKILL.md with BOM to verify scanner robustness
Use Cases
When conducting authorized reverse engineering, vulnerability research, CTF competitions, security audits, or penetration testing, you want the model to automatically route to corresponding specialized skills based on task type (e.g., IDA reverse engineering, APK unpacking, JS signature location, AD attack chain reproduction). Also suitable for importing the entire upstream reverse-skill methodology into the dsh workflow at once, avoiding the need to copy SKILL.md files one by one.
Prerequisites & Compatibility
| Dependency | Minimum Version | Description |
|---|---|---|
| @deepseek-ai/cordis | ^4.0.1 | Cordis framework (peerDependency, provided by dsh runtime) |
| @deepseek-ai/dsh-skill | ^0.0.1-rc.1 | dsh skill seam types (peerDependency, provided by dsh runtime) |
| Node.js | Not declared (repo CI uses Node 22, @types/node ^22.0.0) | Source uses ES2022 / NodeNext |
| Platform | Cross-platform | Only uses Node built-in modules (fs, url, path), no native dependencies |
| Native Modules | None | No native bindings like node-pty, node:sqlite |
Installation
dsh plugin --profile web add github:dhicoc/dsh-reverse-skill
Configuration
This plugin requires no additional configuration. All behavior is determined by the directory contents of skills/ and CTF-Sandbox-Orchestrator/ in the repo — no schema, no runtime parameters.
FAQ
Q: After installation, I'd like to confirm the skills are actually registered. Is there a command to list all 85 skills?
A: You can run npm install && npm test in the repo root. This command will recompile and assert through the actual registered SkillProvider that all 85 skills can be discovered, have no duplicate names, and return non-empty content from get().
Q: I only want a subset of skills. Can I pick them out individually?
A: The plugin uses directories as the data source and cannot select subsets at runtime. If you only want to use certain skills, you can copy the corresponding subdirectories to the customSkillDirs path and mount them as a preset (the non-plugin fallback solution provided in the README).
Q: If SKILL.md contains allowed-tools / disallowed-tools, will they be enforced?
A: No. dsh currently treats these as unknown fields and delays execution; tool constraints declared in the skill body need to be ensured by yourself at the harness layer.
Q: Where did the upstream agents/*.yaml (OpenAI Agents SDK) configurations go?
A: They've been removed. dsh's ctx.subagent only supports spawning Codex / Claude Code CLI, and cannot map OpenAI Agents SDK definitions; these 43 agent definitions were deliberately not included in the plugin.
Q: Will external MCP servers referenced in skills (burp-mcp, IDA MCP, jshookmcp, etc.) start automatically?
A: No. The plugin only packages SKILL.md and necessary local resources. For external MCP servers, please self-integrate and start them according to dsh's mcp.servers configuration.
Q: Will uninstalling this plugin affect dsh itself?
A: No. This plugin only registers a provider on ctx.skills. After uninstallation, that provider stops exposing skills, and dsh's core functionality is unaffected.
Q: Can I use this on Windows?
A: Yes. The source code only uses Node built-in modules, and port.py is only used during the搬运 phase (directories already on disk don't need to run it again), so it's cross-platform at runtime.
Learning Curve
Beginner — one-line command to install and use, no schema to fill in, no runtime parameters; prerequisite is understanding dsh's own skill invocation process and knowing which scenarios fall within authorized security research.
Known Issues & Limitations
allowed-tools/disallowed-toolsare not enforced by dsh and must be ensured at the harness layer- camelCase frontmatter fields (like
when_to_use) have been batch rewritten towhenToUsebyport.py; when manually modifying and appending new skills, maintain consistency yourself - Upstream
agents/*.yamlOpenAI Agents SDK definitions were not ported, as dsh does not support that Agent framework - External MCP servers referenced in skill body (burp-mcp, IDA MCP, etc.) are not within the plugin distribution scope and need to be self-integrated
- Documentation links have been rewritten to
../and../../CTF-Sandbox-Orchestrator/to adapt to dsh's flat mount path
reverse-skill 的完整 DeepSeek Harness(dsh)插件版。 把上游
zhaoxuya520/reverse-skill(25k★,MIT)全部 85 个 SKILL.md 原样封装成一个 dsh Cordis 插件,随包分发、随插件加载,无需手动维护候选清单。
这是什么
- 完整移植:85 个 skills =
skills/下 43 个(60+ 领域技能经去重后的真实 SKILL.md 数)+CTF-Sandbox-Orchestrator/下 42 个(CTF 赛道技能)。与上游一一对应,不裁剪、不挑捡。 - 插件形态(你选的):以 dsh 一等公民的
skillseam(ctx.skills)注册一个 provider,harness 启动时自动把全部技能注入可用技能库。 - 对比已有不完全移植:社区里的
dsh-reverse-security只移植了 45 个、且是 preset-only(无 Cordis 插件)。本仓库补齐到 85 个并提供正式插件入口。
适用范围(请遵守)
本仓库内容仅用于 授权的 逆向工程、渗透测试与安全研究。使用者须确保对目标系统拥有合法授权。一切未授权行为与本仓库无关。
目录结构
dsh-reverse-skill/
├── src/
│ └── index.ts # 数据驱动的 Cordis 插件:递归扫描并注册全部 SKILL.md
├── skills/ # 43 个领域技能(上游 skills/ 1:1 复制)
│ ├── reverse-skill-router/ # 路由技能(由上游 skills/SKILL.md 重构成目录)
│ ├── pentest-tools/ # 含嵌套子技能 src-hunter 等
│ └── reverse-engineering/ # 含嵌套子技能 dsl-vm-reverse 等
├── CTF-Sandbox-Orchestrator/ # 42 个 CTF 赛道技能
├── port.py # 归一化脚本(上游 → 本仓库的搬运/前导matter修正)
├── package.json
├── tsconfig.json
└── LICENSE # MIT(与上游一致)
相对路径:
src/index.ts通过fileURLToPath(new URL('../skills', import.meta.url))定位资源(编译后lib/index.js同样成立,因为skills/与lib/同属包根目录的相邻子目录)。注意这里不能再套一层dirname()——../skills这个 URL 已经指向目录本身,多套dirname会把它截断成包根目录,从而把node_modules里的 SKILL.md 也算进来。
安装(插件形态)
1. 安装依赖与构建
# 安装 peer 依赖(cordis / dsh-skill 由 dsh 运行时提供,这里用于类型与构建)
npm install
npm run build # tsc → 生成 lib/ 与 lib/types/
package.json 中已声明:
"main": "lib/index.js",
"types": "lib/types/index.d.ts",
"peerDependencies": {
"@deepseek-ai/cordis": "^4.0.1",
"@deepseek-ai/dsh-skill": "^0.0.1-rc.1"
}
2. 在 dsh 中启用本插件
本仓库已声明 dsh.bundle manifest(见 cordis.patch.yml),因此可直接用一行命令安装并激活:
# 从 GitHub 安装并激活(推荐)
dsh plugin add github:dhicoc/dsh-reverse-skill
安装后 dsh 会读取 cordis.patch.yml 把 reverse-skill 这个 Cordis 插件插入当前 profile,启动时自动注册 85 个技能。若你想在 profile / package 配置里手动引用,包名是 @dhicoc/dsh-reverse-skill:
# dsh 配置(示例,键名可能因版本而异)
plugins:
- "@dhicoc/dsh-reverse-skill"
加载后,插件在 apply(ctx) 里调用 ctx.skills.registerProvider(...),把 85 个技能注册进 ctx.skills。模型可通过 ctx.skills → tool-skill 自动调用,用户也可通过技能名手动调用(受各 SKILL.md 的 user-invocable 控制)。
3. (可选)非插件回退:直接当 preset 用
本仓库同时携带完整的 skills/ 与 CTF-Sandbox-Orchestrator/ 目录,可作为 preset 直接挂载,无需构建:
skills:
local:
customSkillDirs:
- "./dsh-reverse-skill/skills"
- "./dsh-reverse-skill/CTF-Sandbox-Orchestrator"
dsh 技能发现优先级(先命中先生效):项目
.dsh→ 项目.agents→customSkillDirs→ 用户.dsh→ 用户.agents。扁平发现,不接受递归**/SKILL.md,所以路由技能必须是reverse-skill-router/SKILL.md这样的目录结构(本仓库已处理好)。
插件工作原理(数据驱动,零手写清单)
src/index.ts 不做任何硬编码候选列表,而是:
- 递归遍历
skills/与CTF-Sandbox-Orchestrator/,找到每个SKILL.md; - 解析前导 matter(含把
metadata.user-invocable提升为顶层user-invocable、when_to_use→whenToUse的归一化); - 构造
SkillCandidate(含resourceBase: {kind:'directory', path}、结果缓存); - 注册一个
SkillProvider,get()时返回完整 body。
新增/删除技能只需改目录,插件自动同步。
可复跑验证
在仓库根目录运行:
npm test
该命令会重新编译插件,并通过实际注册的 SkillProvider 断言 85 个已打包技能都能被 list() 发现、名称无重复且均能按需 get() 返回非空正文。测试还会临时创建一个带 UTF-8 BOM 和 CRLF 的 SKILL.md,确认扫描器不会静默跳过此类文件;fixture 在测试结束后会自动删除。发布工作流也会在 npm publish 前运行同一检查。
已知限制(诚实告知)
agents/*.yaml不可移植:上游 43 个 OpenAI Agents SDK 的 agent 定义无法映射到 dsh 的ctx.subagent(dsh 仅支持拉起 Codex / Claude Code CLI)。这些 agent 定义未纳入插件。allowed-tools/disallowed-tools不被 dsh 强制:dsh 当前把这两项视为未知字段,延迟执行。技能内的工具约束需自行在 harness 层保证。- camelCase 前导 matter 字段(如
when_to_use)会被 dsh 拒绝:port.py已统一修正为whenToUse、user-invocable等受支持字段。 - MCP 工具(如 burp-mcp)需另行配置:技能正文里引用的外部 MCP server 不在本插件范围内,请按 dsh 的
mcp.servers自行接入。 - 文档链接已重写:正文内相对链接已改为
../(及 CTF 相关为../../CTF-Sandbox-Orchestrator/),以适配 dsh 扁平挂载路径。
License
LICENSE 为 MIT,与上游 reverse-skill 保持一致。内容版权归原 upstream 作者与贡献者;本仓库为 dsh 适配封装。
Read the usage guide →
Install steps, key points, FAQ and compatibility for this plugin — auto-derived from indexed fields.
Listing badge
[](https://deepseek-plugin.org/plugins/dhicoc/dsh-reverse-skill)Paste this markdown into your GitHub README to link back to this listing. The badge only states the listing — not a security endorsement.