drivelineresearch/autoresearch-claude-code

This skill turns Claude Code into an autonomous experiment runner. It sets up a loop that tries ideas, logs results, commits or reverts changes, and regenerates a dashboard—all without human intervention. The loop only stops when a real budget boundary (max runs, wall-clock time, or target metric) is hit or the user explicitly ends it. It helps ML engineers, researchers, and AI agent developers who need to optimize a metric across hundreds of runs without sitting at the terminal.