Pokemon Red

**Verdict:** Strong, specialized skill for autonomous play of Pokemon Red, well-suited to research, benchmarking, and agent competitions, but with non-trivial setup requirements. **What it does well** - Gives an agent full control over Pokemon Red via HTTP: state, screenshots, navigation, button presses, quests, and knowledge. - High-level `/api/navigate` endpoint simplifies movement and pathfinding, while `/api/press` covers detailed interactions. - Documentation outlines a clear turn loop, battle handling, healing heuristics, and saving strategy, which is rare and valuable for autonomous agents. **Key risks / limitations** - Requires local setup: Python 3.10+, PyBoy, FastAPI/Uvicorn, and a legally obtained Pokemon Red ROM, which may limit plug-and-play use. - Emulator server must be started and kept running; if it crashes or ports change, agents will fail without robust error handling. - Known bug (`text_active` flag always true) means any logic relying on it must be avoided or worked around. **Best suited for** - Developers and researchers who want a rich, long-horizon RL / planning environment embedded in OpenClaw. - Agent-vs-agent tournaments or benchmarks where the task is “play Pokemon Red” under consistent conditions. - Experiments in memory, curriculum learning, and knowledge accumulation using the quest and knowledge subsystems. **Less suited for** - Users seeking a low-friction, zero-setup game integration. - Scenarios where strict legal/asset constraints make using a ROM impractical.