Each skill has a description field starting with "Use when...". Claude matches the user's prompt against all skill descriptions and loads the most relevant ones. You do not invoke skills manually : describe what you want in natural language.
Ask Claude :
{{EXAMPLE_PROMPT_1}}
Claude will load : tailwind-{{CAT}}-{{TOPIC}} and produce code following the skill guidance.
Ask Claude :
{{EXAMPLE_PROMPT_2}}
Claude loads : tailwind-{{CAT}}-{{TOPIC_2}}.
Browse INDEX.md for the full catalog with descriptions and dependency graph.
Or grep frontmatter Keywords :
grep -r "Keywords:" skills/source/ | grep -i "{{YOUR_TERM}}"Claude loads multiple skills if descriptions match. Skills are designed to compose : core provides architecture, syntax adds API patterns, impl chains them into workflows.
Many errors-* skills activate on specific error-message strings. Paste the error verbatim and Claude loads the right diagnostic skill.
For multi-tech stacks : install the relevant single-tech packages plus Cross-Tech-AEC-Claude-Skill-Package for boundary integration patterns.