Skill UI

This tool provides a read-only view over per-turn decision logs, aggregating data on skill co-occurrence patterns, meta-skill usage statistics, and router fixtures. It returns a structured JSON summary, making it ideal for analyzing historical skill usage, optimizing workflows, and feeding insights into downstream LLM models.

This skill scaffolds a gold-standard annotation project tailored for evaluating clinical NER and de-identification models (OpenMed). It guides users through defining label schemas, writing rigorous guidelines, configuring annotation tools (like BRAT/Label Studio), generating synthetic clinical text, and creating disciplined, non-leaking train/dev/test splits. It ensures that all committed data are synthetic fixtures for reproducible evaluation.