claude-autoresearch-skill

Run `npx skills add AlexWortega/claude-autoresearch-skill --all -g -y` in your terminal to install the skill globally.

It researches existing solutions for your ML task, plans a diverse experiment matrix, and runs a long generational loop of bounded train-and-eval experiments until it finds the best verified config.

No. It asks where to get compute first (Kaggle, local GPU, or cloud SSH), and if none is reachable it falls back to design-only mode with a runnable experiment harness.

A peer-critic panel prunes weak or duplicate hypotheses before any training starts, and every kept winner is adversarially re-verified against a held-out metric.

Yes. The generational loop runs as a background workflow and keeps proposing and testing ideas until it hits your budget, stagnation limit, or convergence.