Select a scenario to run, or a run / eval to inspect the evidence. Drive the agent from your terminal (Claude Code in this project) — its edits, runs, and world changes appear here live.