Minecraft Agents — AI skill for Claude Code
Six autonomous agents playing Minecraft together, driven by a local LLM, with.
How to install Minecraft Agents
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open bjcoombs/minecraft-agents and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Minecraft Agents does
Six autonomous agents playing Minecraft together, driven by a local LLM, with.
Alternatives in AI
- Rb Setup — First-time setup 1.3k ★
- Ralphex — Extended Ralph loop for autonomous AI-driven plan execution 1.1k ★
- DontFeedTheAI — Transparent anonymization proxy for AI-assisted pentesting 654 ★
README
minecraft-agents
Six autonomous agents playing Minecraft together, driven by a local LLM, with persistent memory that survives between sessions.
Built over one evening. This repo exists so the learnings outlive the chat context they were discovered in.
What it does
Six mineflayer bots — `Claude`, `Woodcutter`, `Builder`, `Miner`, `Forager`, `Fighter` — each an independent Node process with:
- A local LLM brain (
llama3.1:8bvia Ollama) deciding what to say and do - Persistent memory — a per-agent wiki compiled nightly from a raw journal
- A shared objective — an 11-stage progression toward beating the game
- Roles — gathering, mining, crafting, foraging, combat
- Improvised dialogue — catchphrases and combat shouts written in the moment
Everything runs locally. No API keys, no cost, no internet at runtime.
What actually happened
Reached **7 of 11 stages**: wood → tools → stone → iron → iron pickaxe → diamonds → obsidian. Six of those seven were genuinely earned, including crafting a diamond pickaxe unaided after a placement bug was fixed. The Nether was never reached.
Along the way they mined at y=-54, smelted their own iron, fished, farmed wheat, fought zombies, kept chests, slept in beds, and wrote 210 wiki entries about their own days — including this unprompted self-criticism from Woodcutter:
*Stored 10 items and then 46 items in a single chest. This is not efficient. Consider investing in more chests or improving inventory management.*
Layout
Follows [Karpathy's LLM-wiki pattern](https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f) — immutable evidence, compiled understanding, and the schema that connects them.
CLAUDE.md instructions for agents working on this repo
SCHEMA.md conventions (layer 3)
wiki/ compiled learnings (layer 2)
bugs.md every bug, root cause, and the log line that proved it
architecture.md how it fits together; the concurrency traps
models.md
Related Skills
Agentmachi
A Hamachi server for agents. Start a hub, get an address, agents join and work together — like Hamachi and pla
Tinyagent TS
TinyAgent brings the core ideas of the original Python library into a lean, zero-dependency TypeScript package
Bullpen Coach
Switch Sage (the wellness coach) on or off. When on, Sage sends proactive check-ins (rest reminders, ship cele
CEGM
Live LLM-driven Cheat Engine layer with MCP HTTP and a local dashboard for single-player offline experiments
Prompt Atlas
Auditable, citation-driven LLM prompt review skill across 9 frontier vendors and 2-9B small-local models
Singularmem
Singularmem is a local-first persistent memory layer for LLM-driven workflows built in Rust, SQLite, Typescrip
Related Agents
Brain Retrieval Engineer
Retrieval engineer — six retrieval strategies (hybrid, graph-enhanced vector, entity-anchored local, guarded T
Task Reviewer
Strict quality reviewer for implementation task files. Validates completeness, self-containment, and clarity f
Exploiter
Autonomous agent for crafting and testing LLM exploitation payloads