User Guide

A practical walkthrough for using EurekaClaw — from installation to reading your first generated paper.

• Input Modes
• Browser UI
• Understanding the Output
• Gate Mode and Human Review
• Tuning for Cost vs. Thoroughness
• Skills and Continual Learning
• Troubleshooting

---

At a Glance

EurekaClaw has three input modes depending on how specific your goal is:

Mode	Command	When to use
Prove	eurekaclaw prove "<conjecture>"	You have a precise mathematical statement
From Papers	eurekaclaw from-papers <ids>	You want to extend specific papers
Explore	eurekaclaw explore "<domain>"	You want to discover open problems

After a run, artifacts land in ./results/<session_id>/ — LaTeX source, compiled PDF, proof state, and bibliography.

Common Options

Option	Default	Description
--domain, -d	""	Research domain (enables domain-specific tools and skills)
--output, -o	./results	Output directory
--gate	none	none / auto / human — how much to pause for review
--mode	skills_only	Post-run learning: skills_only, rl, madmax
--verbose, -v	—	Enable DEBUG logging