Skip to content

INSTALL.md

Latest commit

 

History

History
64 lines (48 loc) · 1.79 KB

INSTALL.md

File metadata and controls

64 lines (48 loc) · 1.79 KB
Raw

Installing Litmus

Via ClawHub (recommended)

Ask your OpenClaw agent:

"Install the litmus skill"

Or from the CLI:

npx clawhub install litmus

Manual install

git clone https://github.com/kuberwastaken/litmus ~/.openclaw/skills/litmus

Requirements

  • GPU: NVIDIA GPU with CUDA (required for training runs)
  • uv: Python package manager — curl -LsSf https://astral.sh/uv/install.sh | sh
  • git: For experiment version control and git worktrees
  • python3: For JSON attempt records, leaderboard scripts, and analytics
  • nvidia-smi: Recommended for GPU detection (usually bundled with CUDA drivers)
  • OS: Linux or macOS (Windows not supported)

First-time data setup

After installing the skill, run setup once to download training data (~1 GB):

bash {baseDir}/scripts/setup.sh

This clones Karpathy's autoresearch training harness, installs Python deps via uv, prepares the dataset, and creates the shared lab git repository at ~/.litmus/repo/. Each agent will get its own branch in this shared repo via git worktrees.

Takes ~5 minutes on a good connection.

What setup creates

~/.litmus/
  repo/          ← shared lab git repo (all agents' experiment branches live here)
  harness/       ← training code (cloned from autoresearch)
  shared/
    attempts/    ← one JSON file per experiment (source of truth for leaderboard)
    notes/
      discoveries/
      anomalies/
      moonshots/
      synthesis/
    skills/      ← validated reusable techniques (Skills Library)
  agents/        ← per-agent git worktrees (created by prepare-agents.sh)

Uninstall

npx clawhub uninstall litmus
rm -rf ~/.litmus   # removes all runtime state, agent workspaces, and results