AI Workflow Skills: Agent Skills for AI Training and Fine-Tuning

AI Workflow Skills is a collection of reusable Agent Skills for AI training and fine-tuning, maintained by Taiyi AI Lab.

Each skill lives in its own folder and uses SKILL.md to describe when it should activate and how it should run, with reference docs, scripts, and templates. It currently includes grpo-finetune (end-to-end local GRPO fine-tuning on CUDA: environment checks, reward functions, training scripts, dry runs, evaluation, and troubleshooting) and mint-lora-training (cloud MinT LoRA fine-tuning on macaron.im: SFT, DPO, GRPO, evaluation, adapter download, and quota checks).

Install by downloading pinned .zip files from GitHub Releases into your agent skill directory (e.g. ~/.codex/skills/), or add Taiyi-AI-Lab/AI-Workflow-Skills via the Claude Code plugin marketplace. The repo includes npm run validate to check manifests, build release zips, and scan for secrets or private local paths.