reverse-skill: Block AI agent reverse engineering tool selection confusion with 41 routing rules

Resolves the confusion that arises when AI agents such as Claude Code, Codex, and Cursor encounter APKs, binaries, frontend JS encryption, CTF challenges, or penetration testing targets. It prevents the problem of agents guessing commands because they cannot determine which tool to use among jadx, apktool, Frida, IDA, and BurpSuite. Instead, this package routes to the appropriate methodology based on the input task type, verifies available tools, and then executes reproducible workflows.

Supports a wide variety of scenarios including APK analysis, iOS mobile reverse engineering, ELF/EXE binary analysis, .NET/C# decompilation, JS encryption parameter analysis, DSL VM, HTTP capture, malware analysis, penetration testing, red team orchestration, CTF, firmware/IoT, patch diff, Pwn, EDR bypass, API/GraphQL, supply chain security, LLM/AI security, and OLLVM devirtualization. Each scenario is separated into a dedicated skill module and validated with 41 routing rules (R0-R40) and 163 regression benchmark cases. It supports Windows and Ubuntu and maintains a client-neutral structure that is not dependent on specific AI clients.

Unlike existing approaches, it solves the problem of tools, MCP servers, and scripts being scattered across multiple machines and the same mistakes being repeated. A single structured configuration file (routing.json) serves as the single source of truth for the routing core and is verified through cross-platform CI. It is useful for security researchers, CTF players, or developers building AI-based security automation when delegating complex reverse engineering and penetration testing tasks to AI agents. It is distributed under the MIT license, while some submodules such as CTF-Sandbox-Orchestrator follow the GPLv3 license, and Pentest Swarm AI follows the AGPL-3.0 license.