skill-creator

To create a new skill, you draft a SKILL.md file with accurate YAML frontmatter containing the name and description fields, then use validation workflows to verify the directory structure and package the skill components.

Progressive disclosure structures skill instructions so only essential context loads initially, revealing deeper references or scripts as needed. This minimizes context bloat and ensures Claude triggers the correct behavior without overwhelming the session.

Yes, you can package scripts, references, and assets together within a skill directory. The validation workflow checks these optional components and ensures the final zip package maintains the correct structure for local-first harness deployment.

To fix a skill that fails to trigger, inspect session failures for evidence, validate the YAML frontmatter name and description fields for accuracy, and iteratively update the instruction context to align with the expected tool workflow.

Yes, validating and packaging a skill requires Python dependencies such as yaml for parsing frontmatter and zipfile for packaging the directory structure, alongside sys, pathlib, and re for validation workflows.