Auto Research

Autonomous experiment loop: modify code, evaluate, measure metric, keep wins, discard losses. Inspired by Karpathy's autoresearch.

Works in any project with a measurable metric — ML training, compiler optimization, algorithm tuning, performance benchmarking.

Discovery Flow

If research.toml exists, skip to Setup Phase.

Otherwise:

1. Explore the project silently — read README/docs, scan file structure, identify language/framework, find existing benchmarks or evaluation scripts

2. Ask the user one question (use AskUserQuestion if available):