zhaoxuya520/reverse-skill

What it solves

AI agents often struggle with cybersecurity tasks because they don't know which specialized tools (like jadx, Frida, or IDA Pro) to use for specific file types or scenarios, and they often repeat the same mistakes. This project provides a structured routing system that guides AI agents through the correct methodology and toolchain for a given security task, replacing guesswork with repeatable workflows.

How it works

The project acts as a "skills router" for AI agents. It uses a central configuration (routing.json) containing 43 routing rules to map a user's task to a specific methodology. The workflow follows a strict sequence: it checks global rules, uses a primary routing ladder to identify the correct skill module, initializes the case scope, and then executes the scenario-specific skill using the appropriate tools, MCP servers, or scripts. The system is client-neutral, meaning it can be integrated into various AI clients like Claude Code, Cursor, or OpenCode.

Who it’s for

Security researchers, penetration testers, and CTF competitors who use AI agents to assist in reverse engineering, malware analysis, and vulnerability research.

Highlights

• Comprehensive Scenario Coverage: Supports a wide array of tasks including APK/iOS analysis, binary reverse engineering, JS deobfuscation, firmware pentesting, and LLM security.
• Rigorous Validation: Includes a regression benchmark with 173 cases to ensure routing accuracy and a coherence check for supply-chain security.
• Tool-Agnostic Integration: Provides a platform-neutral core that works across Windows, Ubuntu, and Kali Linux.
• Structured Evidence Chain: Implements a case-init and field-journal system to track evidence, findings, and paths for professional reporting.