Autoresearch · arjunkmrm bundle Autonomous experimentation loop. Given a codebase and a metric, run bounded experiments autonomously — modify code, measure, keep or revert, repeat. Inspired by Karpathy's autoresearch: agent modifies code within constraints, evaluates via a single metric, logs everything. Agent collaborates with the human to write the strategy (program), then executes the loop. Use for: performance optimization, bundle size reduction, latency tuning, algorithm improvement, architecture exploration, or any open-ended 'make this number better' task. Triggers on: 'autoresearch', 'experiment loop', 'optimize this metric', 'run experiments overnight', 'autonomous optimization', 'tune this', 'make this faster/smaller/better'.