Dataset Transformation

When you're prepping datasets for ML training or evaluation and need to convert between formats like OpenAI chat, SageMaker SFT/DPO, HuggingFace preference, or Bedrock Nova, this generates the transformation code for you instead of making you write it inline. It walks through an 11-step workflow that determines whether you're doing training or eval data (this matters because the schema resolution is different), figures out your source and target formats, examines samples, then generates and refines a Python transformation function before running it. Works with local files or S3. The workflow is thorough but rigid: you go step by step, confirm at each stage, and it won't let you skip around.

Install to Claude Code

npx -y skills add awslabs/agent-plugins --skill dataset-transformation --agent claude-code