Back to directory

PS2 Recomp Mastery

An autonomous Claude Code skill that packages MIPS R5900 architecture knowledge and PS2Recomp project conventions so an AI agent can reverse-engineer and recompile PlayStation 2 games with persistent memory across sessions.

PS2 Recomp Mastery

This is a skill for AI coding agents such as Claude Code, Antigravity, or Cursor that gives them architectural knowledge, a procedural workflow, and persistent memory for reverse-engineering and recompiling PlayStation 2 games through the PS2Recomp project. It is meant to be used with the agent treated as an autonomous junior reverse engineer that runs cmake builds and the game executable itself, rather than as a chatbot the human drives step by step.

The agent keeps a PS2_PROJECT_STATE.md file as persistent memory, so a session can be paused and resumed later without losing track of which registers or functions were being analyzed, and it has circuit breakers that stop it and ask for help after repeated failures on the same crash rather than burning tokens indefinitely.

Using it requires a working native build toolchain (Visual Studio with Clang and CMake components), Ghidra with the EmotionEngine Reloaded plugin and a GhydraMCP extension for the agent to drive Ghidra directly, and a custom PCSX2 debug build.

More in Agent skills

See all 581
  • Provenance gate

    github.com

    A deterministic authorization gateway that sits between an AI agent and its tools, tracing where each parameter of a side-effect call came from, and blocking the call or asking a human when the source does not check out.

  • A markdown framework of system rules that restates Isaac Asimov's Four Laws of Robotics as a set of prioritized loss calculations for large-language-model agents to follow, published as a discovery file for agents and crawlers to read.

  • Vitruvius

    github.com

    An engineering research agent that runs a discover, read, synthesize, verify and review loop over technical questions, reading sources directly rather than inferring from titles, and publishing its own scored verifier benchmark, weaknesses included.