pokemon-player

Plays Pokemon games via headless PyBoy emulation with RAM state reads and screenshot verification.

Updated Sep 10, 2026

Search Tags:#automation #pokemon #gaming #gameplay #emulator #pyboy #gameboy

One-click install

npx skills add https://github.com/loteiron/ZeusAgent --skill pokemon-player-loteiron

Or copy as Structured Prompt for Agent▼

Please help me install this Agent Skill.
Skill: pokemon-player
Source: https://github.com/loteiron/ZeusAgent/tree/main/optional-skills/gaming/pokemon-player
Command: npx skills add https://github.com/loteiron/ZeusAgent --skill pokemon-player-loteiron

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires pokemon-agent, pyboy.

What problem does it solve? Playing a Game Boy Pokemon game autonomously requires bridging emulator control, game state awareness, and visual navigation, which is difficult to coordinate manually through trial-and-error inputs. ## Core Features & Use Cases - Headless Emulation Server: Runs the pokemon-agent package with PyBoy to serve game state, screenshots, and action endpoints over HTTP on port 9876. - Observe-Decide-Act Loop: Combines RAM state reads (position, HP, battle, dialog) with vision-analyzed screenshots to navigate maps, handle battles, and advance dialog. - Save State Management: Creates named save states before risky encounters and resumes sessions via API or startup flags. - Use Case: A user says "play pokemon" and the agent sets up the emulator, exposes a live dashboard through a localhost.run tunnel, and plays through Pokemon Red gym by gym while recording progress to memory. ## Quick Start Say "play pokemon" and provide your own legally obtained Game Boy ROM file so the agent can start the emulator and begin playing.

Frequently Asked Questions about pokemon-player

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

Why does the agent get lost while navigating in the game?▼

RAM state only reports position and HP, not surroundings like ledges, doors, or NPCs. The skill mitigates this by taking screenshots every 2-4 movement steps and using vision analysis to verify position after each action sequence.