creating-skills

To turn code patterns into reusable AI instructions, extract them into a SKILL.md file containing YAML frontmatter, a generic pattern layer, and a source layer with real-code file references. This format teaches the assistant when and how to apply the pattern.

Engineering patterns for AI code modification require a SKILL.md format with YAML frontmatter defining the name and description, followed by a two-layer structure: a generic pattern section and an 'In the source' block showing real code with file path comments.

To prevent incorrect application of a coding pattern, validate the SKILL.md entry against a six-test quality checklist: stranger, trigger, real-code, before/after, over-application, and domain connection tests. This ensures the assistant triggers the pattern only when appropriate.

The best way to structure examples for AI skill authoring is using realistic before/after code transformations that include WHY comments. This demonstrates the exact code changes required and explains the reasoning, allowing the AI to apply the modification safely.

Yes, you need real source code references to create a transferable AI skill. The format requires a grounded source section containing actual code and file path comments to ensure the pattern represents operational knowledge rather than ad-hoc notes.

You should convert a repo's error-handling pattern into a SKILL.md file when the pattern recurs across multiple projects and you need it to automatically load as operational knowledge when prompts ask for improved error handling in developer workflows.