reverse-skill
Reverse Engineering / Authorized Penetration Testing / Security Research Skill Router Pack AI-powered routing + On-demand toolchain bootstrapping + Self-evolving knowledge base Supports Claude Code, Kiro, Cursor, Cline, and other AI coding clients 逆向/渗透/安全技能路由包 - AI 自动路由 + 按需自举工具链 + 自动进化经验库 | 支持 Claude Code / Kiro / Cursor / Cline 等代码 AI 客户端
- 评测生成时间(北京时间)
- 本报告引擎
- v3.0.0
- 当前引擎
- v3.17.3
本报告与当前引擎使用不同规则;原分数不会自动更新,不同版本的分数不宜直接对比。
进入后确认来源与额度,提交才会创建任务。
代码版本未记录或记录无效,无法核对本报告的 commit 与送评文本;不会用当前项目代码回填历史记录。
综合采用结论
检测到关键风险,修复前不建议接入
历史报告未保存置信度明细;重新评测后可查看各项证据贡献。
本报告评了什么
综合分用于静态采用筛选,不是任务成功率,也不代表使用 Skill 后的能力提升。
- 静态检查范围
- 记录 25 个文件 · 97,145 字符
- 只检查采集到的受限文本与已知风险模式;未读取的脚本、依赖和运行环境不在结论内。
- AI 内容复核
- 未使用
- 复核送评文本的说明和设计,不是安装测试或实际任务执行。记录缺失时不推断已使用或未使用。
- 真实任务效果
- 本报告未提供实测证据
- 没有使用/不使用 Skill 的任务对照、重复运行或耗时与 token 成本结果。采用前仍需在隔离环境验证自己的任务。
脚本引用覆盖
脚本引用覆盖记录不可得;不推断没有引用或已经检查。
本次报告未完成图示提取
本次评测未完成 AI 证据提取;确定性评分与安全扫描仍然有效。重新评测后,证据充分时会自动选择合适图型。
- 问题与用途描述
- 有效 README
- 安装或接入步骤
- 可执行示例
- 发现提示词或系统信息提取意图
- 缺少输出或结果说明
- 缺少错误处理或排障
受控试用与进一步验证
生产环境与敏感数据场景
也有自己的公开项目?先免登录查已有报告;没有报告时,可保留地址继续登录评测。
查找我项目的报告 →prompt-extractionCTF-Sandbox-Orchestrator/competition-forensic-timeline/SKILL.md:18high confidence3. Build one minimal chain from foothold to persistence, execution, access, or exfiltration.修复:移除提示词提取逻辑,并增加敏感上下文不可输出的边界说明。
prompt-extractionCTF-Sandbox-Orchestrator/competition-mailbox-abuse/SKILL.md:3high confidencedescription: Internal downstream skill for ctf-sandbox-orchestrator. CTF-sandbox workflow for enterprise mail abuse, OAuth consent, inbox or forwarding rules, transport rules, shar修复:移除提示词提取逻辑,并增加敏感上下文不可输出的边界说明。
prompt-extractionCTF-Sandbox-Orchestrator/competition-mailbox-abuse/SKILL.md:20high confidence5. Reproduce the smallest mail effect that proves persistence, exfiltration, or privilege.修复:移除提示词提取逻辑,并增加敏感上下文不可输出的边界说明。
prompt-extractionCTF-Sandbox-Orchestrator/competition-mailbox-abuse/SKILL.md:50high confidence- The smallest replayable sequence that proves persistence, exfiltration, or delegate access修复:移除提示词提取逻辑,并增加敏感上下文不可输出的边界说明。
- 01移除提示词提取逻辑,并增加敏感上下文不可输出的边界说明。
- 02补充输出或结果说明
- 03补充错误处理或排障
方法、证据与局限展开收起
GitHub Repository API
25 个文件 · 97,145 字符
v3.0.0 · AI 内容复核:未使用
- 静态评测不会安装或执行项目代码
- 安全扫描基于高信号文件与已知模式,不能替代人工审计
- 流行度只反映采用程度,不代表安全或工程质量
30 天热度趋势
README
reverse-skill
Cybersecurity Skills Router · 逆向技能路由包
Navigate the dark waters, sail against the stream.
About · Getting Started · Usage · Tutorial · Fast route · Routing · Ops contracts · AI Bootstrap · Sponsors · Contributing
🌐 中文 · Project website · Online tutorial
About
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
| Routing rules | Regression benchmark | Core skill modules | CI platforms | Client model |
|---|---|---|---|---|
| 44 (R0–R45) | 175 cases | 45 tracked modules | Windows + Ubuntu | Client-neutral |
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/
Sponsors
AstraFlow | UCloud AstraFlow provides one-click access to 200+ leading open-source models, including Kimi K3, DeepSeek V4/V3, Qwen 3, GLM 5.2, and HappyHorse—no model training required, ready to use out of the box. |
|
| Atlas Cloud is a full-modal AI inference platform that provides unified API access to 400+ curated image, video, audio, 3D, and language models. Atlas Cloud supports reverse-skill with model services for cross-platform routing verification, documentation, and open security workflows. |
|
| Kite AI builds identity and payment infrastructure for the agent economy. Its support helps sustain reverse-skill's open-source maintenance, routing benchmarks, and platform-neutral security workflows. |
Built With
IDA Pro · radare2 · Ghidra · Binary Ninja
Getting Started
Prerequisites
- 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
git clone https://github.com/zhaoxuya520/reverse-skill.git
Then refresh the tool index per platform:
| Platform | Command |
|---|---|
| Windows | powershell -File skills/scripts/refresh-tool-index.ps1 |
| Linux / macOS | bash skills/scripts/refresh-tool-index.sh |
| Kali Linux | bash kali/scripts/refresh-tool-index.sh |
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
Usage
Supported scenarios
| Scenario | Entry |
|---|---|
| APK / Android analysis | skills/apk-reverse/ |
| iOS / mobile | skills/mobile-reverse/ |
| Binary reverse (exe/dll/so/elf) | skills/ida-reverse/ / skills/radare2/ |
| Binary Ninja / HLIL / MLIL / MCP | skills/binary-ninja-reverse/ |
| .NET / C# | skills/dotnet-reverse/ |
| Frontend JS / encrypted params | skills/js-reverse/ |
| DSL VM / custom JS opcode VM | skills/reverse-engineering/dsl-vm-reverse/ |
| HTTP capture / request replay | anything-analyzer, Reqable MCP + js-reverse/ |
| Malware / YARA | skills/malware-analysis/ |
| Penetration testing / scanning | skills/pentest-tools/ |
| Attack chain / red-team orchestration | skills/attack-chain/ |
| Case evidence review / report handoff | skills/case-review/ |
| CTF competition | CTF-Sandbox-Orchestrator/ (42 sub-skills) |
| Firmware / IoT | skills/firmware-pentest/ |
| Patch diff / N-day | skills/patch-diff-exploit/ |
| Pwn / exploit development | skills/pwn-chain/ |
| EDR bypass | skills/edr-bypass-re/ |
| API / GraphQL | skills/api-security/ |
| Supply chain / SBOM | skills/supply-chain-security/ |
| LLM / AI security | skills/llm-security/ |
| OLLVM deobfuscation | skills/reverse-engineering/references/ollvm-deobfuscation.md |
| Diagrams / reports | skills/diagram-generator/ / skills/docs-generator/ |
Key files
| File | Purpose |
|---|---|
| README_AI.md | AI agent bootstrap and configuration |
| RULES.md | Global routing rules (scope gate before ACT) |
| skills/MASTER-ROUTING.md | PRIMARY fast ladder |
| skills/routing.md | Task → skill routing matrix |
| skills/SKILL.md | Master entry point |
| skills/INDEX.md | Auto-generated, client-neutral skill navigation index |
| skills/config/routing.json | Routing single source of truth (43 rules, R0–R44) |
| skills/tool-index.md | Local tool status (auto-generated) |
| skills/scripts/master-route.ps1 | One-shot PRIMARY triage (reads routing.json) |
| skills/scripts/case-init.ps1 | Case dir: scope / timeline / workitems |
| skills/case-review/ | Read-only Evidence graph review and artifact fixity checks |
| skills/scripts/test-routing.ps1 | Routing regression runner (173 benchmark cases) |
| skills/scripts/verify-routing-coherence.ps1 | Structure + supply-chain pin gate checks |
| skills/scripts/extract-summaries.ps1 | Regenerates INDEX.md from skill frontmatter |
| AGENTS.md | Platform-neutral repository instructions |
| skills/ops/ | Scope, Evidence chain, roles, timeline (skill-router form) |
Testing (run after any routing/config change)
# 1. Routing regression — 173 (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
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.
For Codex, the repository also exposes an optional adapter plugin at plugins/reverse-skill/. It delegates to the repository's existing routing core and does not register external MCP servers automatically.
Repository layout
.
├── 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)
Contributing
Contributions are welcome! Fork the repo, create a feature branch, and open a PR.
- Fork the Project
git checkout -b feature/AmazingFeaturegit commit -m 'Add some AmazingFeature'git push origin feature/AmazingFeature- Open a Pull Request
Contributors
License
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
Acknowledgments
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.
Contact
- Email: ww7517437@gmail.com
- Issues: GitHub Issues
Disclaimer
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
Installation and download security
See Security Policy, Installation and Download Security Guidance, and the 2026-09-03 repository security review.
Community quick start and issue triage
See Quick Start and Community Issue Triage for installation, client integration, and how community issues are classified. Installation/archive security details remain in Installation and Download Security Guidance.