Modify machine learning training code and evaluate models in a keep/discard loop.
npx skills add https://github.com/chrishuffman5/skills --skill autoresearch-chrishuffman5
Please help me install this Agent Skill. Skill: autoresearch Source: https://github.com/chrishuffman5/skills/tree/main/autoresearch Command: npx skills add https://github.com/chrishuffman5/skills --skill autoresearch-chrishuffman5
Autoresearch solves the problem of manual experimentation and optimization in machine learning training. It allows you to run autonomous research experiments that modify training code, train models, evaluate results, and iterate — all without human intervention.
/loop feature for hands-off experimentation.Use Claude Code's /loop feature to start an autonomous research experiment:
/loop 7m run the next autoresearch experiment — check results.tsv for history, modify train.py with a new idea, train, evaluate, keep or discard, log to results.tsv
High-intent search queries and answers about installing and using this skill.
You can automate machine learning experiments by using an autonomous loop that modifies training code, trains models, evaluates results, and keeps or discards changes based on performance metrics.
The keep or discard loop evaluates model training results after each autonomous code modification, keeping changes only if they improve model performance and discarding them otherwise to iteratively optimize training.
Yes, you can run autonomous ML research without manual intervention by integrating with Claude Code's /loop feature to continuously execute experimental cycles of modifying code, training, and evaluating models.
You need a train.py file for model training code and a results.tsv file to log experiment history before starting the autonomous model optimization loop to track modifications and outcomes.
Automated code modification provides effective model optimization by autonomously applying new training ideas and validating them through a keep or discard loop, reducing manual experimentation overhead.
Autonomous experimentation limitations include dependency on the initial training code quality and the need for reliable evaluation metrics to accurately decide whether to keep or discard model training modifications.
ACautomataInspect a codebase and generate a Claude Code workflow for ML experiments.
tbroadleyExtract samples, messages, events, and metrics from Inspect AI .eval logs.
ChimeraFoundationaDelegate coding tasks to Claude Code via Hermes CLI with PTY for interactive sessions.
pluginagentmarketplaceOptimize machine learning models with quantization, pruning, and ONNX export.