Build with an AI agent

picogame ships two things that make AI coding agents (like Claude Code) genuinely good at building games for it: a ready-made skill that teaches the agent how to design and implement a picogame game, and LLM-readable documentation it can pull in wholesale.

The picogame game-design skill

picogame-game-design is an Agent Skill — a packaged set of instructions and references an agent loads when you ask it to make a game. It carries:

Game-design fundamentals — core loop, game feel & juice, difficulty & fairness, scope discipline, so the result is actually fun, not just running.
The engine, mapped — every building block and what it costs in RAM, plus the full API reference (exact signatures for the native C engine and every helper library).
Genre playbooks — Breakout, shmup, platformer, racing, first-person raycaster dungeon, and more — each with the core loop, controls, real tuning numbers, pitfalls and an MVP.
Technique recipes — state machines, enemy AI, collision, procedural generation, pseudo-3D (Mode-7 & raycasting), each mapped to picogame.
A runnable starter game and a desktop-simulator workflow, so the agent builds and checks with screenshots without any hardware.

Install it

Download and unzip it into your agent’s skills folder:

Download the skill (.zip)

For Claude Code that folder is ~/.claude/skills/:

cd ~/.claude/skills
unzip ~/Downloads/picogame-game-design-skill.zip

Then just ask — “make a tiny shooter for picogame” — and the skill loads automatically.

The skill’s source lives in the public repo under skills/.

LLM-readable docs (llms.txt)

For agents that read documentation directly, the whole site is available as clean markdown following the llms.txt convention — point your agent at these instead of scraping HTML:

/llms.txt — the index
/llms-full.txt — the entire documentation as one markdown file
/_llms-txt/api.txt — just the API (reference + engine + helpers), for writing code
/_llms-txt/getting-started.txt — the intro, tutorials and concepts

Editing a game.json with an agent

A level made in the web editor is one text file, game.json, that the agent edits directly — map rows over the tileset’s legend, zones and their story data, effects — next to story.py (Python story scripts) and code.py. The rules the skill teaches:

keep the legend’s characters (append, never re-letter), so the map’s diff stays a picture;
run python3 tools/scene_build.py check and fmt before handing the file back, art after touching a PNG; never edit .pal8 files or build/;
with a folder chosen in the editor, the agent’s write shows up there within two seconds; in git, one map row is one line, so your edits and the agent’s merge cleanly.