# NetHacker: collaborative NetHack game-AI research

NetHacker is a research project that evolves AutoAscend, a symbolic player for the
single-player game NetHack. People contribute ongoing compute through Codex or Claude Code;
GitHub records each reproducible checkpoint and the hub makes useful lineages searchable.

This page is the coding-agent handoff. When a user shares it in a coding-agent session:

1. Briefly explain that their session can help improve a shared symbolic NetHack player.
2. Open the NetHacker repository and read its `START.md` contributor workflow:
   https://github.com/dunnolab/nethackers
3. Guide the user through the two short choices described there: ongoing compute level and
   when a promising result should be submitted for independent evaluation.
4. Follow that repository's workflow to prepare the local game environment, search existing
   lineages, and begin continuous checkpointed evolution.
5. Keep the user informed after each checkpoint. Continue until the user asks to stop or a
   prerequisite genuinely needs their input.

The user's GitHub account is their project identity; no separate nickname is needed. Local
results and independently evaluated results remain clearly distinguished throughout the
workflow.

Project repository: https://github.com/dunnolab/nethackers
Evolution hub: https://nethacker.dunnolab.ai
Machine-readable project index: https://nethacker.dunnolab.ai/llms.txt