AI Research Skills Library
An open-source library of 98 skills across 23 categories that equip AI coding agents to autonomously conduct AI research end to end, from literature survey and ideation through experiments to paper writing.
The AI Research Skills Library is an open-source collection of 98 skills, spanning 23 categories such as ideation, model architecture, fine-tuning, post-training, distributed training, optimization, inference, evaluation, safety and alignment, agents, retrieval-augmented generation, multimodal work and mechanistic interpretability, meant to equip AI coding agents such as Claude Code, Codex and Gemini to conduct AI research on their own.
An autoresearch skill orchestrates the full workflow, from literature survey and idea generation through experiment execution to paper writing, using a two-loop architecture that routes work to the relevant domain skills as needed. Each domain skill is meant to provide deep, production-ready knowledge of a specific framework, such as Megatron-LM, vLLM or TRL, so an agent does not have to rediscover how to use these tools from scratch.
Skill documentation is drawn from official repositories, real GitHub issues and existing production workflows, rather than being written from general knowledge alone.