skill-creator

To create and evaluate Claude skills, draft a SKILL.md with clear triggering descriptions, populate evals/evals.json, and run iterative evaluation loops. This structured workflow enables data-backed confidence by tracking improvements across train/test splits over multiple iterations.

The best way to structure skill descriptions for triggering is through descriptor optimization. Refine your SKILL.md descriptions iteratively to improve user discovery and ensure the skill activates accurately based on specific query intents.

To package validated skills into deployable files, run quick validation scripts to check structure and formatting, then compile the SKILL.md, evals.json, and supporting scripts into a deployable .skill file for production environments.

Yes, you can manage train and test splits for skill evaluation workflows. The evaluation orchestration feature manages splits, runs multiple prompts per query, and aggregates results to provide data-backed confidence for deployment decisions.

Yes, PyYAML is required to automate skill lifecycle workflows. It serves as the core dependency for parsing and managing configuration files like SKILL.md and evals.json during the creation, evaluation, and packaging processes.