File: README.md Copy Chapter

reverse-skill

Cybersecurity Skills Router · 逆向技能路由包

About Copy Chapter

If you are an AI Agent, jump to README_AI.md and follow the instructions strictly.

When an AI agent (Claude Code, Codex, Cursor, OpenCode, or another compatible client) encounters an APK, a binary, frontend JS encryption, a CTF challenge, or a pentesting target, this package routes it to the right methodology, checks available tools, and executes a repeatable workflow instead of guessing commands.

User task
  → RULES.md
  → MASTER-ROUTING / master-route.ps1 (PRIMARY)
  → case-init / scope.md (auth + network_profile; no target ACT until ready)
  → Scenario skill → tools / MCP / scripts
  → timeline + Evidence→Finding→Path → report + field-journal

Why this exists:

• AI agents don't know whether to use jadx, apktool, Frida, IDA, or BurpSuite for a given task
• APK, ELF, JS, PCAP, and CTF tasks each need different playbooks
• Tools, MCP servers, and scripts are scattered across machines
• The same mistakes get repeated because experience isn't reused

Current status Copy Section

The routing core is driven by one structured configuration, validated by cross-platform CI, and kept separate from optional client adapters.

PRIMARY ladder: skills/MASTER-ROUTING.md · Full matrix: skills/routing.md · Ops: skills/ops/

(back to top)

Built With Copy Section

IDA Pro · radare2 · Ghidra

(back to top)

Getting Started Copy Chapter

Prerequisites Copy Section

• Java / JDK — for jadx and apktool
• Node.js 22.12+ — for JS toolchain and MCP servers
• Python 3.x — for Frida and helper scripts
• A code AI client — Claude Code, Codex, Cursor, OpenCode, or another compatible client

Installation Copy Section

git clone https://github.com/zhaoxuya520/reverse-skill.git

Then refresh the tool index per platform:

Check skills/tool-index.md to see detected tools.

Platform-specific docs:

• Kali Linux → kali/README-kali.md
• Ubuntu/Debian → docs/platforms/linux.md
• macOS → docs/platforms/macos.md

(back to top)

Usage Copy Chapter

Supported scenarios Copy Section

Key files Copy Section

Testing (run after any routing/config change) Copy Section

# 1. Routing regression — 163 (hint → expected PRIMARY) cases, fails CI on any mismatch
powershell -NoProfile -ExecutionPolicy Bypass -File skills/scripts/test-routing.ps1
# 2. Structure coherence + supply-chain pin gate (unpinned auto-install fails)
powershell -NoProfile -ExecutionPolicy Bypass -File skills/scripts/verify-routing-coherence.ps1
# 3. Smoke: verify + script parse + quick route matrix
powershell -NoProfile -ExecutionPolicy Bypass -File skills/scripts/smoke.ps1
# 4. INDEX.md drift check (regenerate with extract-summaries.ps1 if dirty)
powershell -NoProfile -ExecutionPolicy Bypass -File skills/scripts/extract-summaries.ps1 -Check

GitHub Actions CI runs all of the above on Windows + Ubuntu for every push/PR.

Client-neutral integration Copy Section

The routing core, regression suite, manifests, and case workflow do not depend on a specific AI client. Claude Code, Codex, Cursor, OpenCode, and other clients should load the repository through their own adapter or project-instruction mechanism. Client-specific configuration must remain optional and outside the core routing contract.

Repository layout Copy Section

.
├── README.md / README_zh.md / README_AI.md
├── RULES.md / RULES_zh.md
├── skills/
│   ├── MASTER-ROUTING.md / SKILL.md / routing.md
│   ├── ops/                   # ops contracts
│   ├── scripts/               # master-route, case-init, bootstrap, verify
│   ├── field-journal/
│   ├── apk-reverse/ mobile-reverse/ js-reverse/ dotnet-reverse/
│   ├── ida-reverse/ radare2/ reverse-engineering/ malware-analysis/
│   ├── pentest-tools/ attack-chain/ pwn-chain/ firmware-pentest/
│   ├── api-security/ supply-chain-security/ llm-security/
│   └── ...
├── CTF-Sandbox-Orchestrator/
├── docs/
├── kali/                      # see kali/README-kali.md
└── work/                      # local cases (gitignored)

(back to top)

Sponsors Copy Chapter

(back to top)

Contributing Copy Chapter

Contributions are welcome! Fork the repo, create a feature branch, and open a PR.

1. Fork the Project
2. git checkout -b feature/AmazingFeature
3. git commit -m 'Add some AmazingFeature'
4. git push origin feature/AmazingFeature
5. Open a Pull Request

Contributors Copy Section

(back to top)

License Copy Chapter

This project (reverse-skill) is primarily licensed under the MIT License (see LICENSE).

Submodule and third-party dependencies:

• CTF-Sandbox-Orchestrator/: GNU GPLv3
• Pentest Swarm AI: Original project is AGPL-3.0. This repo only invokes it via CLI or MCP and does not include its source code
• Other tools (jadx, frida, nmap, burpsuite-mcp, etc.) are subject to their respective official licenses

(back to top)

Acknowledgments Copy Chapter

Thanks to all open-source tool authors. This project integrates tools across reverse engineering, penetration testing, CTF, and security analysis — every tool is the fruit of community effort.

Special thanks to the OLLVM deobfuscation ecosystem contributors and everyone who submitted test samples, issues, and PRs.

(back to top)

Contact Copy Chapter

• Email: [email protected]
• Issues: GitHub Issues

Disclaimer Copy Chapter

This project is intended solely for lawful security research, education, CTF competitions, and testing of systems that you own or have explicit authorization to assess.

Unauthorized access, scanning, exploitation, disruption, data acquisition, or any other use against systems without prior permission is strictly prohibited. Users are solely responsible for complying with applicable laws, regulations, and the authorized scope of testing. The maintainers accept no liability for misuse of this project or for any resulting damage or legal consequences

File: burp-mcp-full/README.md Copy Chapter

BurpSuite MCP Full Control Extension

通过 MCP 协议完整控制 BurpSuite 的所有核心功能。跨平台支持 Windows / Linux (Kali) / macOS。

快速开始 Copy Chapter

1. 编译扩展 Copy Section

Windows:

cd burp-mcp-full
build.bat

Linux / Kali / macOS:

cd burp-mcp-full
chmod +x build.sh
./build.sh

构建脚本会自动：检测 JDK 21+、下载依赖（montoya-api 2025.5 / gson / nanohttpd）、编译、把扩展描述符（META-INF/extensions/burp-extension.properties）打入 jar、打包 fat jar。无需 Gradle。

输出：build/libs/burp-mcp-full.jar。

2. 加载到 Burp Copy Section

Burp Suite → Extensions → Add → Java → 选择 build/libs/burp-mcp-full.jar

加载后在 Output 看到：

[MCP] Server started on http://127.0.0.1:9876

3. 鉴权（v2 起默认启用） Copy Section

扩展启动时自动生成随机 token 并写入 ~/.burp-mcp-token。mcp-bridge.js 会自动读取该文件并在每个请求携带 Authorization: Bearer <token> 头，无需手动配置。

需要固定 token 时（例如多个客户端共享），可用：

• JVM 参数：-Dburp.mcp.token=<token>
• 环境变量：BURP_MCP_TOKEN=<token>（同时用于 bridge 侧）

所有 /health、/tools、/（POST）请求均要求携带该头，否则返回 403。CORS 已收敛为仅允许 http://127.0.0.1 来源。

4. 配置 MCP 客户端 Copy Section

在任何 MCP 客户端（Claude Code / Kiro / Cursor / Cline / Windsurf）中添加（stdio 模式）：

{
  "mcpServers": {
    "burpsuite": {
      "command": "node",
      "args": ["<本目录路径>/mcp-bridge.js"]
    }
  }
}

5. 开始使用 Copy Section

对 AI 说："分析 Burp 代理历史中的请求，找出安全漏洞"

功能列表 Copy Chapter

扩展暴露 78 个工具。常用分类如下（完整列表见 src/main/java/com/burpmcp/McpHttpServer.java 的 getToolList()，或访问 GET http://127.0.0.1:9876/tools，需携带 Authorization 头）：

扫描/爬取（scan、scan_active、crawl）需要 Burp Professional。Community 版会返回明确的许可证错误。手动添加的 issue（add_issue）会写入 Site map。

关键工具参数 Copy Chapter

`intruder_attack` — 自动化枚举攻击 Copy Section

`scan` — 启动审计 Copy Section

启动后用 scan_results 轮询 issues 与活动审计状态（请求数、错误数、插入点数）。

`register_proxy_rule` — 代理请求拦截规则 Copy Section

通过 remove_proxy_rule 注销规则（基于 Registration.deregister()，真正从 Burp 卸载）。

调用示例 Copy Chapter

查看代理历史 Copy Section

POST http://127.0.0.1:9876
{"tool": "proxy_history", "params": {"limit": 10, "url_filter": "personalblog"}}

发送请求 Copy Section

POST http://127.0.0.1:9876
{"tool": "send_request", "params": {"method": "GET", "url": "https://example.com/api/test"}}

自动化枚举攻击（核心功能） Copy Section

POST http://127.0.0.1:9876
{
  "tool": "intruder_attack",
  "params": {
    "url_template": "https://target.com/api/verify?code=@@",
    "method": "POST",
    "from": 0,
    "to": 999999,
    "pad_digits": 6,
    "success_length_not": 176,
    "headers": {"User-Agent": "Mozilla/5.0"}
  }
}

开关拦截 Copy Section

POST http://127.0.0.1:9876
{"tool": "intercept_toggle", "params": {"enable": false}}

端口配置 Copy Chapter

默认监听 127.0.0.1:9876。如需更改（例如与 PortSwigger 官方 MCP 扩展同端口冲突）：

1. Burp 侧：启动 Burp 时传 JVM 参数 -Dburp.mcp.port=9877，或设环境变量 BURP_MCP_PORT=9877。
2. 桥接侧：MCP 客户端配置里设环境变量 BURP_MCP_PORT=9877 与 BURP_MCP_HOST=127.0.0.1。

两侧端口必须一致。若 Burp 未运行或端口不通，桥接会在 tools/list 与 tools/call 返回明确的连接错误指引。

故障排查 Copy Chapter

源码构建（Gradle 可选） Copy Chapter

cd burp-mcp-full
gradle jar      # 需本机已装 Gradle 8.7+
# 输出: build/libs/burp-mcp-full.jar

推荐使用 build.bat / build.sh（零依赖，自动下载 jar）。Gradle 路径仅作备选。

File: CTF-Sandbox-Orchestrator/README.md Copy Chapter

CTF Sandbox Orchestrator

一个面向 Codex / Skills 体系的竞赛沙盒技能集合。

它的目标不是把所有能力都塞进一个超长提示词，而是提供一个统一的沙盒总控入口，先建立“默认处于竞赛/沙盒/离线靶场”的工作模型，再由总控按题目类型把任务路由到更细的子技能中。

项目定位 Copy Chapter

这个仓库主要用于解决以下场景：

• CTF
• AWD / 攻防演练
• 本地离线靶场
• 沙盒化漏洞分析
• Web / API / Cloud / Container / Windows / AD / Reverse / Pwn / DFIR / Crypto / Mobile / AI Agent 等混合型题目

核心思路：

• 默认把用户提供的目标、域名、节点、身份、二进制、日志、流量与附件视为竞赛沙盒内部资产
• 优先建立最小可验证路径，而不是一开始就泛化分析
• 由一个总控 skill 统一编排，再按主导证据面切到子 skill
• 子技能只做下游专项，不抢总控入口

核心设计 Copy Chapter

1. 单一入口 Copy Section

默认入口是：

• ctf-sandbox-orchestrator

它负责：

• 建立沙盒假设
• 选择最合适的分析路径
• 控制上下文膨胀
• 在需要时调用子技能

2. 子技能下游化 Copy Section

所有 competition-* 技能都被设计为 downstream-only：

• 不应在未激活总控的情况下隐式触发
• 应由 ctf-sandbox-orchestrator 主动路由调用
• 每次只加载当前最相关的专项能力，避免无关技能污染上下文

3. 面向多类型竞赛题 Copy Section

当前仓库覆盖了多类技能方向，例如：

• Web 运行时 / 路由 / WebSocket / GraphQL / 文件解析 / 请求归一化
• Prompt Injection / Agent / Cloud / Metadata / K8s / Container Escape
• Reverse / Pwn / Malware / Firmware / PCAP / 自定义协议重放
• Windows / AD / Kerberos / DPAPI / 证书滥用 / Relay / Mailbox
• Android / iOS / Crypto / Stego / Mobile Runtime
• ZIP / PKZIP legacy encryption / bkcrack known-plaintext recovery

仓库结构 Copy Chapter

E:\WorkSpace\competition
├─ ctf-sandbox-orchestrator
├─ competition-web-runtime
├─ competition-agent-cloud
├─ competition-reverse-pwn
├─ competition-identity-windows
├─ competition-prompt-injection
├─ ...
└─ LICENSE

其中：

• ctf-sandbox-orchestrator：总控入口
• competition-*：专项子技能
• references/：总控使用的路由矩阵与领域参考说明
• agents/openai.yaml：各技能的调用约束与入口控制

推荐使用方式 Copy Chapter

方式一：从总控进入 Copy Section

优先激活：

• ctf-sandbox-orchestrator

然后让总控根据题目自动决定下一步，例如：

• Web 题路由到 competition-web-runtime
• 容器 / 云题路由到 competition-agent-cloud 或更细粒度子技能
• Windows / AD 题路由到 competition-identity-windows
• 二进制 / 崩溃 / 恶意样本题路由到 competition-reverse-pwn

方式二：保留总控，按需下钻 Copy Section

当已经确认主导证据面后，由总控继续下钻到具体子技能，而不是让用户手动切换整个工作模型。这样可以保持：

• 沙盒假设一致
• 输出风格一致
• 路由策略一致
• 子技能职责清晰

致谢 Copy Chapter

本项目已在 LINUX DO 社区 发布，感谢社区的支持与反馈。

File: docs/platforms/linux.md Copy Chapter

Generic Linux Setup

This guide covers ordinary Linux distributions such as Ubuntu, Debian, Linux Mint, Pop!_OS, and other Debian-family systems. Kali Linux has a dedicated path: ../../kali/README-kali.md.

Positioning Copy Chapter

The core knowledge layer is platform-agnostic. On Linux, what changes is the execution layer:

• package manager: usually apt, sometimes distro-specific package managers;
• Python isolation: prefer pipx or a venv because modern systems may enforce PEP 668;
• GUI apps: BurpSuite, IDA Pro, Ghidra paths differ by installation method;
• security tools: distro packages may be old or absent, so GitHub releases are often preferred.

Quick setup baseline Copy Chapter

sudo apt update
sudo apt install -y \
  git curl wget ca-certificates unzip tar jq \
  python3 python3-venv python3-pip pipx \
  openjdk-17-jdk nodejs npm \
  graphviz plantuml nmap sqlmap ffuf hashcat binwalk

python3 -m pipx ensurepath

Ubuntu repository versions can be old. For Node.js, radare2, jadx, nuclei, and Ghidra, prefer the alternatives below when the distro package is missing or outdated.

Tool installation matrix Copy Chapter

Recommended path layout Copy Chapter

~/tools/
├── jadx/
├── ghidra/
├── SecLists/
├── android-platform-tools/
└── anything-analyzer/

~/.local/bin/              # pipx / user scripts
/opt/idapro/               # optional IDA Pro install
/usr/bin/                  # distro packages

Installing common tools Copy Chapter

jadx from GitHub release Copy Section

mkdir -p ~/tools/jadx
curl -L https://github.com/skylot/jadx/releases/latest/download/jadx-1.5.5.zip -o /tmp/jadx.zip
unzip -q /tmp/jadx.zip -d ~/tools/jadx
export PATH="$HOME/tools/jadx/bin:$PATH"
jadx --version

If the release filename changes, download the latest Linux/ZIP asset from https://github.com/skylot/jadx/releases.

Frida via pipx Copy Section

pipx install frida-tools
frida --version
frida-ps --version

radare2 Copy Section

Prefer the official release or source install when the distro package is absent:

git clone https://github.com/radareorg/radare2 ~/tools/radare2-src
cd ~/tools/radare2-src
sys/install.sh
r2 -v

Ghidra Copy Section

mkdir -p ~/tools/ghidra
# Download latest release from https://github.com/NationalSecurityAgency/ghidra/releases
# unzip ghidra_*.zip into ~/tools/ghidra

Android platform-tools alternative Copy Section

mkdir -p ~/tools/android-platform-tools
# Download platform-tools-latest-linux.zip from Android Developers
# unzip into ~/tools/android-platform-tools
export PATH="$HOME/tools/android-platform-tools/platform-tools:$PATH"
adb version

MCP setup notes Copy Chapter

The core bootstrap is client-neutral by default. It does not write ~/.claude/mcp.json or ~/.codex/config.toml unless an MCP host is explicitly selected. For MCP capabilities, use one of:

--mcp-host=claude
--mcp-host=codex
--mcp-host=both

Without an explicit host, the bootstrap may prepare the runtime but reports the MCP capability as registration-required. Override the explicit adapter paths with CLAUDE_MCP_CONFIG or CODEX_CONFIG_PATH when needed.

BurpSuite MCP Copy Section

Build the extension:

cd burp-mcp-full
chmod +x build.sh
./build.sh

Then load the generated jar in BurpSuite:

Burp Suite → Extensions → Add → Java → build/libs/burp-mcp-full.jar

MCP stdio bridge example:

{
  "mcpServers": {
    "burpsuite": {
      "command": "node",
      "args": ["/absolute/path/to/reverse-skill/burp-mcp-full/mcp-bridge.js"]
    }
  }
}

jshookmcp Copy Section

{
  "mcpServers": {
    "jshook": {
      "command": "npx",
      "args": ["-y", "@jshookmcp/[email protected]"],
      "env": {
        "JSHOOK_BASE_PROFILE": "search"
      }
    }
  }
}

Bootstrap capabilities and refresh tool index Copy Chapter

From the repository root, list all bootstrap capabilities:

bash skills/scripts/bootstrap-reverse.sh --list

Install ordinary capabilities without selecting an Agent client:

bash skills/scripts/bootstrap-reverse.sh jadx apktool frida

For MCP registration, select the target client explicitly:

bash skills/scripts/bootstrap-reverse.sh jshookmcp --mcp-host=codex
bash skills/scripts/bootstrap-reverse.sh anything-analyzer --mcp-host=claude
bash skills/scripts/bootstrap-reverse.sh idapro --start-services --mcp-host=both

Refresh the tool index only:

bash skills/scripts/bootstrap-reverse.sh --list
bash skills/scripts/refresh-tool-index.sh

This generates:

skills/tool-index.md
skills/tool-index.json

The bootstrap script installs or prepares supported capabilities where possible using the same core capability names as Windows. MCP client registration is performed only when --mcp-host=... is explicit. The refresh script detects common Linux/macOS tools and discovers supported MCP registrations without choosing a default Agent client. Manual-only tools such as IDA Pro and BurpSuite still require local installation and app-specific setup.

Validation checklist Copy Chapter

java -version
python3 --version
node -v
npm -v
npx -v
jadx --version || true
apktool --version || true
adb version || true
frida --version || true
r2 -v || true
ghidraRun 2>/dev/null || true
bash skills/scripts/refresh-tool-index.sh

When to use Kali docs instead Copy Chapter

Use ../../kali/README-kali.md when:

• the host is actually Kali;
• you want Kali-native security tooling and MCP integrations;
• you need Metasploit, NetExec, responder, BloodHound, Certipy, HexStrike, or other offensive-security distributions pre-wired.

For Ubuntu / Debian, keep this generic Linux guide as the default and only borrow Kali scripts when the tool is known to exist on your system.

File: docs/platforms/macos.md Copy Chapter

macOS Setup

This guide covers macOS users who want to use Cybersecurity Skills Router with Claude Code, Codex CLI, Cursor, Cline, Windsurf, Kiro, or another Agent client.

Positioning Copy Chapter

The core Skill layer is platform-agnostic. On macOS, the main differences are:

• package manager: Homebrew is preferred;
• Python tools: use pipx or venv rather than global pip;
• GUI apps: BurpSuite, IDA Pro, and Ghidra may live under /Applications;
• Android tooling: android-platform-tools is available through Homebrew;
• some Linux security tools may need GitHub releases, Go, or manual setup.

Quick setup baseline Copy Chapter

/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

brew install \
  git curl wget jq unzip gnu-tar \
  python node openjdk \
  jadx apktool android-platform-tools \
  radare2 graphviz plantuml \
  nmap sqlmap ffuf hashcat binwalk

python3 -m pip install --user pipx
python3 -m pipx ensurepath

Homebrew package names can change over time. If a formula is missing, use the alternative path in the tool matrix below.

Tool installation matrix Copy Chapter

Recommended path layout Copy Chapter

~/tools/
├── SecLists/
├── anything-analyzer/
└── custom-releases/

/Applications/
├── Burp Suite*.app
├── IDA Professional*.app
└── Ghidra*.app

/opt/homebrew/bin/        # Apple Silicon Homebrew
/usr/local/bin/           # Intel Homebrew
~/.local/bin/             # pipx / user scripts

Installing common tools Copy Chapter

Frida via pipx Copy Section

pipx install frida-tools
frida --version
frida-ps --version

SecLists Copy Section

mkdir -p ~/tools
git clone https://github.com/danielmiessler/SecLists ~/tools/SecLists

anything-analyzer Copy Section

mkdir -p ~/tools
git clone https://github.com/Mouseww/anything-analyzer ~/tools/anything-analyzer
cd ~/tools/anything-analyzer
corepack enable
pnpm install
pnpm dev

If the service uses a custom port or token, update your Agent client's MCP configuration accordingly.

MCP setup notes Copy Chapter

The core bootstrap is client-neutral by default. It does not write ~/.claude/mcp.json or ~/.codex/config.toml unless an MCP host is explicitly selected. For MCP capabilities, use one of:

--mcp-host=claude
--mcp-host=codex
--mcp-host=both

Without an explicit host, the bootstrap may prepare the runtime but reports the MCP capability as registration-required. Override the explicit adapter paths with CLAUDE_MCP_CONFIG or CODEX_CONFIG_PATH when needed.

BurpSuite MCP Copy Section

Build the extension:

cd burp-mcp-full
chmod +x build.sh
./build.sh

Load the generated jar in BurpSuite:

Burp Suite → Extensions → Add → Java → build/libs/burp-mcp-full.jar

MCP stdio bridge:

{
  "mcpServers": {
    "burpsuite": {
      "command": "node",
      "args": ["/absolute/path/to/reverse-skill/burp-mcp-full/mcp-bridge.js"]
    }
  }
}

IDA Pro Copy Section

IDA Pro is commercial and must be installed manually. Common locations:

/Applications/IDA Professional.app
/Applications/IDA Free.app
/Applications/IDA Pro *.app

If you use IDA MCP, document the actual app path in your client rules or local environment. Do not hard-code another user's path.

jshookmcp Copy Section

{
  "mcpServers": {
    "jshook": {
      "command": "npx",
      "args": ["-y", "@jshookmcp/[email protected]"],
      "env": {
        "JSHOOK_BASE_PROFILE": "search"
      }
    }
  }
}

Bootstrap capabilities and refresh tool index Copy Chapter

From the repository root, list the same core capability names as the Windows PowerShell bootstrap:

/bin/bash skills/scripts/bootstrap-reverse.sh --list

The generic bootstrap is compatible with the system /bin/bash shipped by macOS (Bash 3.2); Homebrew Bash is not required.

Install ordinary capabilities without selecting an Agent client:

/bin/bash skills/scripts/bootstrap-reverse.sh jadx apktool frida

For MCP registration, select the target client explicitly:

/bin/bash skills/scripts/bootstrap-reverse.sh jshookmcp --mcp-host=codex
/bin/bash skills/scripts/bootstrap-reverse.sh anything-analyzer --mcp-host=claude
/bin/bash skills/scripts/bootstrap-reverse.sh burpsuite-mcp --mcp-host=both

Refresh the local tool index only:

/bin/bash skills/scripts/refresh-tool-index.sh

This writes:

skills/tool-index.md
skills/tool-index.json

bootstrap-reverse.sh installs or prepares supported capabilities where possible on macOS using Homebrew, pipx, npm, and GitHub releases. MCP client registration occurs only when --mcp-host=... is explicit. refresh-tool-index.sh is detection-only and discovers supported MCP registrations without choosing a default Agent client. Manual-only tools such as IDA Pro and BurpSuite still require local app installation and app-specific setup.

Validation checklist Copy Chapter

java -version
python3 --version
node -v
npm -v
npx -v
jadx --version || true
apktool --version || true
adb version || true
frida --version || true
r2 -v || true
brew list --formula | grep -E 'jadx|apktool|radare2|graphviz|plantuml' || true
bash skills/scripts/refresh-tool-index.sh

macOS caveats Copy Chapter

• GUI app paths vary by edition and version. Do not hard-code IDA or Burp paths unless you verified them locally.
• Some security tools are Linux-first. Prefer Homebrew formulae first, then GitHub releases, then source builds.
• iOS analysis may require additional signing, device, and jailbreak-specific setup; keep those steps in a dedicated mobile reverse Skill rather than this generic platform page.

File: kali/README-kali.md Copy Chapter

Cybersecurity Skills Router — Kali Linux 专供版

本目录是 Kali Linux 2026.1 优化适配层。基于 2026 年 3 月发布的 Kali 2026.1（内核 6.18）进行专项优化。
核心知识库（skills/、CTF-Sandbox-Orchestrator/）与 Windows 版共享；Kali 专属 README 和 Bash 入口需要覆盖 Windows 核心能力名，同时额外提供 Kali 原生工具/MCP 能力。
根目录 README-kali.md 仅为短跳转，以本文件为准。

AI：读完后立即执行 Copy Chapter

1. 检测包根目录（含 skills/ 与 kali/ 的仓库根）
2. 读取 kali/RULES-kali.md → 全局注入与工具扫描
3. bash kali/scripts/refresh-tool-index.sh
4. 与主包共用作战链：
   - skills/MASTER-ROUTING.md（或 pwsh skills/scripts/master-route.ps1）
   - skills/scripts/case-init.ps1 → work/<case>/scope.md
   - auth.status=granted + network_profile 后才对目标 ACT
   - skills/ops/（证据链 / 角色 / 时间线 / IDENTITY）
5. 向用户报告配置结果

通用 Agent 引导见仓库根 README_AI.md（检测到 Kali 时再读本文件）。

0. 与 Windows 版的关系（能力名对齐） Copy Chapter

项目根目录/
├── skills/                    # 共享：SKILL、routing、MASTER-ROUTING、ops、scripts、field-journal
├── CTF-Sandbox-Orchestrator/  # 共享：40+ CTF 子技能
├── kali/                      # ← 你在这里
│   ├── scripts/
│   │   ├── bootstrap-reverse.sh
│   │   ├── refresh-tool-index.sh
│   │   ├── bootstrap-manifest.json
│   │   └── lib/
│   │       └── tool-discovery.sh
│   ├── RULES-kali.md
│   └── README-kali.md
├── RULES.md                   # Windows 版规则
└── Readme.md                  # Windows 版说明

0.1 对齐原则 Copy Section

Kali 专属入口不是 Windows README 的简单复制，而是 同一套核心能力名 + Kali 额外能力：

• Windows：skills/scripts/bootstrap-reverse.ps1
• Kali：kali/scripts/bootstrap-reverse.sh
• 普通 Linux/macOS：skills/scripts/bootstrap-reverse.sh

JEB Pro 是用户自行许可和安装的商业工具；Reqable MCP 使用官方固定版本的 reqable-mcp-server，但仍要求单独安装 Reqable 桌面客户端。

Kali 脚本应覆盖 Windows manifest 中的核心能力名，例如 jadx、apktool、frida、jshookmcp、anything-analyzer、idapro、r2、adb、ghidra-mcp、seclists、burpsuite-mcp、nmap、pentestswarm；同时可以额外支持 Kali 原生工具，例如 mcp-kali-server、metasploitmcp、hexstrike-ai、sstimap、xsstrike、netexec 等。

共享的部分（不需要改动）：

• 所有 SKILL.md、routing.md、MASTER-ROUTING.md
• skills/ops/ 作战契约（scope / 证据链 / 角色 / 时间线）
• 所有 references/ 知识库
• field-journal/ 自进化机制
• CTF-Sandbox-Orchestrator/ 全部
• docs-generator/、diagram-generator/
• skills/scripts/case-init.ps1、master-route.ps1（可用 pwsh 调用）

Kali 专属的部分：

• 脚本全部是 bash（.sh）
• 包管理走 apt
• 路径约定为 Linux 风格（/opt/、~/tools/、/usr/bin/）
• 大量工具 Kali 预装，bootstrap 逻辑大幅简化

1. Kali 的天然优势 Copy Chapter

以下工具在 Kali 2026.1 中开箱即用（无需 bootstrap）：

经典预装工具 Copy Section

Kali 2026.1 新增工具（2026年3月） Copy Section

Kali 2025.4 新增工具（2025年12月） Copy Section

Kali 原生 MCP 工具（重点优化） Copy Section

这是 Kali 版相比 Windows 版最大的优势：三个 MCP 工具直接 apt 安装，无需手动配置 GitHub/npm/Docker。

这意味着 bootstrap-reverse.sh 在 Kali 上的工作量远小于 Windows 版。

2. 快速开始 Copy Chapter

2.0 一键初始化（推荐新系统使用） Copy Section

# 全新 Kali 2026.1 系统一键配置（需要 root）
sudo bash kali/scripts/quick-setup.sh

# 跳过系统更新（网络慢时）
sudo bash kali/scripts/quick-setup.sh --skip-update

# 最小安装（不装 AD/内网工具）
sudo bash kali/scripts/quick-setup.sh --minimal

这个脚本会自动完成：系统更新 → 安装 2026.1 新工具 → 配置原生 MCP → 安装逆向工具 → 刷新索引 → 输出报告。

2.1 首次配置 Copy Section

# 1. 进入项目根目录
cd /path/to/cybersecurity-skills-router

# 2. 给脚本加执行权限
chmod +x kali/scripts/*.sh kali/scripts/lib/*.sh

# 3. 刷新工具索引（检测本机工具状态）
bash kali/scripts/refresh-tool-index.sh

# 4. 查看结果
cat skills/tool-index.md

2.2 一键配齐 Kali 原生 MCP（强烈推荐） Copy Section

# 安装 Kali 官方 MCP 三件套
bash kali/scripts/bootstrap-reverse.sh mcp-kali-server metasploitmcp hexstrike-ai

# 安装后 MCP 配置自动写入 ~/.claude/mcp.json
# 如果用 Kiro，手动复制到 ~/.kiro/settings/mcp.json

2.3 安装 2026.1 新工具 Copy Section

# 全部新工具一键安装
bash kali/scripts/bootstrap-reverse.sh adaptixc2 atomic-operator sstimap xsstrike wpprobe fluxion gef

# AD/内网渗透套件
bash kali/scripts/bootstrap-reverse.sh coercer evil-winrm-py netexec responder bloodhound certipy

2.4 安装缺失工具 Copy Section

# 安装单个工具
bash kali/scripts/bootstrap-reverse.sh jadx

# 安装多个工具
bash kali/scripts/bootstrap-reverse.sh jadx apktool frida jshookmcp

# 安装并启动服务
bash kali/scripts/bootstrap-reverse.sh idapro --start-services

2.5 让 AI 客户端自动路由 Copy Section

告诉你的 AI 客户端读取 kali/RULES-kali.md，它会自动完成全局注入。

3. 路径约定 Copy Chapter

4. 与 Windows 版的差异总结 Copy Chapter

5. 验证清单 Copy Chapter

# ─── 基础命令 ───
java -version
python3 --version
pip3 --version
node -v
npx -v

# ─── 逆向工具 ───
jadx --version
apktool --version
adb version
frida --version
r2 -v
gdb --version          # GEF 自动加载

# ─── 渗透工具（Kali 预装） ───
nmap --version
sqlmap --version
hashcat --version
hydra -h | head -1
msfconsole --version
gobuster version
ffuf -V
nuclei -version

# ─── Kali 2026.1 新工具 ───
sstimap -h 2>&1 | head -3
xsstrike -h 2>&1 | head -3
wpprobe --help 2>&1 | head -3
coercer -h 2>&1 | head -3
evil-winrm-py -h 2>&1 | head -3

# ─── AD/内网工具 ───
netexec --help 2>&1 | head -3
responder -h 2>&1 | head -3
certipy --version 2>&1 | head -1

# ─── Kali 原生 MCP ───
which kali-server-mcp && echo "mcp-kali-server OK"
which metasploitmcp && echo "metasploitmcp OK"
which hexstrike-ai && echo "hexstrike-ai OK"

# ─── 刷新工具索引 ───
bash kali/scripts/refresh-tool-index.sh

# ─── 检查 MCP 服务（如果已配置） ───
nc -z 127.0.0.1 5000 && echo "mcp-kali-server OK" || echo "mcp-kali-server offline"
nc -z 127.0.0.1 8085 && echo "metasploitmcp OK" || echo "metasploitmcp offline"
nc -z 127.0.0.1 13337 && echo "IDA MCP OK" || echo "IDA MCP offline"
nc -z 127.0.0.1 23816 && echo "anything-analyzer OK" || echo "anything-analyzer offline"

6. 常见问题 Copy Chapter

Q: Kali 自带的 radare2 版本太旧怎么办？ Copy Section

# 用官方源安装最新版
bash kali/scripts/bootstrap-reverse.sh r2
# Kali 版默认优先 apt 安装/补齐 radare2；如需最新版可按平台文档改用 GitHub/source

Q: 我用的是 Parrot OS / BlackArch，能用吗？ Copy Section

可以。脚本检测的是命令是否存在，不绑定特定发行版。只是 apt 相关的自动安装可能需要改成 pacman（BlackArch）。

Q: IDA Pro Linux 版怎么配？ Copy Section

把 IDA 安装到 /opt/idapro/，然后修改 kali/scripts/bootstrap-manifest.json 中 idapro 的 startScript 路径。

Q: 我想同时在 Windows 和 Kali 上用这套系统 Copy Section

没问题。skills/ 目录通过 Git 同步，field-journal/ 的经验两边共享。只是执行脚本时 Windows 用 skills/scripts/*.ps1，Kali 用 kali/scripts/*.sh。

File: skills/diagram-generator/README.md Copy Chapter

Diagram Generator Skill

This skill helps AI assistants create editable diagrams in Mermaid, Graphviz DOT, PlantUML, and SVG, and optionally render them with local tools.

File: skills/pentest-tools/payloads/README.md Copy Chapter

自定义 Payload 目录

本目录用于存放你自己积累的高效 payload。AI 在执行渗透测试时会优先使用这里的 payload，再去尝试通用 payload。

建议的文件组织 Copy Chapter

payloads/
├── sql-injection/        # SQL 注入 payload
│   ├── mysql-union.txt
│   ├── mssql-stacked.txt
│   └── blind-time.txt
├── xss/                  # XSS payload
│   ├── reflected.txt
│   └── dom-based.txt
├── ssrf/                 # SSRF payload
├── lfi-rfi/              # 文件包含
├── auth-bypass/          # 认证绕过
├── command-injection/    # 命令注入
└── custom/               # 其他自定义

使用规则 Copy Chapter

• AI 测试时先用本目录的 payload，再用 SecLists 通用字典
• 每次发现新的有效 payload，回写到对应目录
• 文件名要有描述性，不要用 1.txt、test.txt

自动补充 Copy Chapter

如果本目录为空，AI 会自动从 SecLists 中提取常用 payload 子集放入。
如果 SecLists 未安装，bootstrap 会自动下载。

File: skills/pentest-tools/src-hunter/README.en.md Copy Chapter

中文 · English

src-hunter

A Claude Code skill for SRC / bug-bounty / pentest workflows. Loads automatically when you mention things like bug bounty, hackerone, waf bypass, or 任意账号, and walks Claude through a five-phase hunt: intake → recon → enum → hunt → report.

It bundles a knowledge base built from public sources: nineteen attack-class playbooks, three hundred-odd structured payloads with WAF/EDR bypass variants, the full HackerOne hacktivity feed (High and Critical, disclosed) at the time of build, and statistical residue from the old WooYun corpus.

Install Copy Chapter

Marketplace:

/plugin marketplace add MyuriKanao/src-hunter-skill
/plugin install src-hunter@src-hunter

Plain git:

git clone https://github.com/MyuriKanao/src-hunter-skill.git ~/.claude/skills/src-hunter

What's in here Copy Chapter

references/
  methodology/    five-phase workflow, attack-priority, bypass toolkit, evidence rules
  playbooks/      one file per attack class — each ends with real H1 cases and a payload library
  industry/       banking/finance and telecom/ISP verticals
  dictionaries/   default credentials and component fingerprints (Chinese stack)
  templates/      submission template (CVSS 4.0)
  h1-reports/     2887 disclosed reports, raw + grouped by weakness
  payloader/      305 structured payloads, 263 WAF/EDR bypass steps, 114 tool cheats

The playbooks are the entry point. They're written from a black-box hunter's perspective — assume you only have a URL — and each one carries the same shape: where to look, what to throw, what response shapes to watch for, how to escalate, and the legal lines.

MCP integration Copy Chapter

The skill wires in a local MCP server as a tool layer so Claude can drive browser automation, CDP debugging, network interception, JS hooks, AST deobfuscation, Frida memory verification, WASM reversing, source-map reconstruction, Android adb bridging, and SSL-pinning bypass directly during the hunt phase.

Current pick: jshookmcp 0.3.0 (134 curated / 386 full / 36 domains). The full index and scenario→tool map lives in references/tools/mcp-jshook.md.

Seven high-affinity playbooks (xss / rce / ssrf-cache-host / mobile / oauth-saml-jwt / api-rest / file-upload) carry a ## 相关 MCP 工具 trailer block telling Claude which jshook tools to reach for and when.

TODO Copy Chapter

• Support more MCP tools
• Multi-agent execution workflow

Triggers Copy Chapter

The skill loads on any of these (and a few more spelled out in SKILL.md):

• bug bounty, hackerone, src 挖洞, 漏洞赏金, 众测
• waf bypass / 绕过 WAF
• "how do I test this endpoint / API / parameter"
• 任意账号 / 任意修改 / 任意删除, 密码重置 / 找回密码
• 默认凭据, Actuator, exposed admin, etc.

Or invoke explicitly: /src-hunter <target>.

Playbook list Copy Chapter

Sources Copy Chapter

• HackerOne hacktivity feed — 2887 disclosed High/Critical reports, fetched directly. Public data.
• WooYun historical archive — 88,636 cases. Only parameter-frequency tables, case IDs, and bypass patterns are retained. The platform is defunct; data is unrecoverable elsewhere.
• Payloader — 305 structured payloads, 263 WAF/EDR bypass steps, 114 tool cheats from the open-source 3516634930/Payloader repo.

This project re-organizes and translates these sources into a hunter-oriented skill. No proprietary data, no scraping of authenticated content.

Rules of engagement Copy Chapter

Each playbook spells out concrete limits. The most-commonly-stepped-on ones:

• Sample size: for SQLi, proving the database name / version is enough — don't dump rows. For IDOR / Mongo / ES exfil, 1–3 samples is the cap.
• Self-account testing: authz bypass, password reset, JWT forgery, redirect_uri abuse, blind XSS — all on two accounts you registered yourself. Never on a stranger's account, even if you can.
• Read, don't write: with RCE, only id / whoami / uname -a. With unauth Redis / Mongo, only info / ping / db.version(). With LFI, stop at root:x: — don't read /etc/shadow.
• No real side effects: no real SMS, no real charges, no real emails, no real refunds, no overwriting files, no editing global announcements / mail templates. Stop once the endpoint returns 200.
• DoS / concurrency: each repro ≤ 60 s, serial, 5 reps total. Race conditions: 50–100 concurrent, never 1000+. Rate-limit bugs: 5–10 hits to your own number, not 1000.
• Don't keep loot: webshells, heapdumps, backups, dumped source — local only, deleted once the report lands. Never push to GitHub or third-party storage.
• Credentials, don't use: leaked AWS / Stripe / DB creds get one validation call (sts get-caller-identity, banner grab) — never spend, send, or connect.
• Redact PII in reports: phone numbers, emails, usernames, tokens, cookies — keep first-2 + last-2, optionally a sha256 fingerprint to prove you had it.
• OOB infrastructure: don't hit public DNSLog services. Use the vendor-provided SSRF testing platform if they offer one, or self-host interactsh / your own DNSLog.
• No capture, no finding: every claim needs an HTTP capture, screenshot, or video. Don't submit "this should work."

The per-playbook rules are tighter (DoS is the most sensitive, file-upload bans webshells, file-read takes one sample). See the last section of each playbook.

License Copy Chapter

MIT. Data sources are public; this project's contribution is the organization, translation to a black-box perspective, and the merge.

File: skills/pentest-tools/src-hunter/README.md Copy Chapter

中文 · English

src-hunter

这是一个给 SRC、众测和 Bug bounty 用的 Claude Code skill。

简单说，就是你给它一个目标，它会按一套固定流程帮你推进漏洞挖掘：先确认目标范围，再做信息收集和资产枚举，然后进入漏洞测试，最后整理报告。

intake → recon → enum → hunt → report

项目内置了一批从公开来源整理的知识库，包括：

• 19 类攻击 playbook
• 305 个结构化 payload
• WAF / EDR 绕过变体
• HackerOne 已披露 High / Critical hacktivity 数据
• WooYun 历史案例统计残余
• 常见国产组件指纹和默认凭据

安装 Copy Chapter

Marketplace：

/plugin marketplace add MyuriKanao/src-hunter-skill
/plugin install src-hunter@src-hunter

Git：

git clone https://github.com/MyuriKanao/src-hunter-skill.git ~/.claude/skills/src-hunter

目录结构 Copy Chapter

references/
  methodology/    五阶段流程、攻击优先级、绕过工具集、证据规则
  playbooks/      每类漏洞一个文件，包含真实 H1 案例和 payload
  industry/       银行/金融、电信/ISP 垂直场景 playbook
  dictionaries/   国产组件指纹和默认凭据
  templates/      CVSS 4.0 报告模板
  h1-reports/     2887 份已披露报告原始数据，并按 weakness 分组
  payloader/      305 个结构化 payload、263 个 WAF/EDR 绕过步骤、114 个工具命令

playbook 是主要入口。所有 playbook 都按黑盒视角编写，默认你只有 URL，没有源码。

每个 playbook 都围绕同一套问题展开：

• 去哪里找入口
• 用什么 payload 测
• 观察哪些响应特征
• 如何判断影响
• 如何提高漏洞价值
• 哪些行为不能做

整体思路不是堆 payload，而是把测试动作、证据留存和报告输出串起来。

MCP 工具集成 Copy Chapter

本 skill 集成本地 MCP 服务器作为工具层，让 Claude 在 hunt 阶段能直接调用浏览器自动化、CDP 调试、网络拦截、JS hook、AST 反混淆、Frida 内存验证、WASM 逆向、Source map 重构、Android adb 桥接、SSL pinning 绕过等能力。

当前主选：jshookmcp 0.3.0（134 工具精选 / 386 全集 / 36 域），完整索引与场景映射见 references/tools/mcp-jshook.md。

7 个高关联 playbook（xss / rce / ssrf-cache-host / mobile / oauth-saml-jwt / api-rest / file-upload）末尾各有 ## 相关 MCP 工具 反向锚点，指明该攻击面下应该调哪些 jshook 工具、何时调。

TODO Copy Chapter

• 支持引入更多 tools
• 多 agent 执行工作流

触发关键词 Copy Chapter

skill 内置触发词包括：

• bug bounty、HackerOne、SRC 挖洞、漏洞赏金、众测
• WAF bypass、绕过 WAF
• 如何测试某个 endpoint / API / 参数
• 任意账号、任意修改、任意删除
• 密码重置、找回密码
• 默认凭据、Actuator、暴露的管理后台

也可以显式调用：

/src-hunter <target>

Playbook 列表 Copy Chapter

数据来源 Copy Chapter

• HackerOne hacktivity feed：2887 份已披露 High / Critical 报告，来源为公开数据。
• WooYun 历史档案：覆盖 88,636 条案例，仅保留参数频率、案例 ID 和 bypass 模式等统计残余。
• Payloader：305 条结构化 payload + 263 个 WAF / EDR 绕过步骤 + 114 条工具命令，原仓库为 3516634930/Payloader。

本项目只整理、翻译和重组公开资料，不包含专有数据，也不抓取需要认证的内容。

红线 Copy Chapter

每个 playbook 末尾都写了具体的边界，下面是抽出来的几个最常踩的点：

• 样本控制：SQLi 探测到库名 / 版本即可证明，不要 dump 数据；IDOR、Mongo / ES 拉数据 1–3 条样本就够，别全量。
• 测试账号自演：越权、密码重置、JWT 伪造、redirect_uri、XSS 盲打全部用自己注册的两个号互测，不要碰陌生人的账号——即使能。
• 只读，不写：拿到 RCE 只跑 id / whoami / uname -a；Redis / Mongo 默认未授权只 info / ping / db.version()；任意文件读看到 root:x: 一行即停，不读 /etc/shadow。
• 不真做副作用动作：不真发短信、不真扣款、不真发邮件、不真退款、不真覆盖文件、不真改公告 / 邮件模板。证明接口能调通 + 200 即停。
• DoS / 并发：单次复现 ≤ 60s，串行做 5 次足够。竞态并发 50–100，绝不 1000+。短信 / 邮件不限速这种，发到自己手机 5–10 次为止。
• 不留物：webshell、heapdump、备份、dump 出来的源码——本地保存，报告后立即删除，不要 push 到 GitHub / 第三方网盘。
• 凭据：拿到不用：泄露的 AWS / Stripe / 数据库凭据，仅 sts get-caller-identity / 看 banner 验证，绝不用来扣款 / 发邮件 / 连接生产库。
• 报告里所有 PII 脱敏：手机号、邮箱、用户名、token、cookie 留前 2 + 后 2，必要时附 sha256 指纹证明拿到过原文。
• OOB 验证：不要使用公开的公共 DNSLog 平台，使用厂商提供的 SSRF 测试平台，或自架 interactsh / 自有 DNSLog。
• 没抓包就没发现：所有断言都要有 HTTP 包 / 截图 / 视频，不要凭"应该"提交。

具体到每类漏洞还有更细的限制（DoS 类最敏感、上传不留 webshell、读类只读 1 条样本等），看对应 playbook 的最后一节。

友情链接 Copy Chapter

linuxdo

License Copy Chapter

MIT。

数据来源均为公开资料。本项目主要做资料整理、翻译、归类，并封装成适合黑盒漏洞挖掘使用的 Claude Code skill。