A lightweight interface for doing research with agents. Currently supports brainstorming ideas with multiple agents in a war room setting to produce a single research proposal. Future work will add experiment agents that carry out proposed experiments, and a master agent that iterates until results satisfy a set of reviewers.
uv syncOpen your favourite agent CLI (Claude, Codex, Gemini, etc.) and chat with:
Run the war room and generate a research proposal on projects/reward-model-interpretability/
Monitor in a second terminal:
python skills/war-room/monitor.py --project projects/reward-model-interpretability/