new-skill

Configure skill auto-activation by adding YAML frontmatter to SKILL.md in `.claude/skills/{skill-name}/`, then register triggers in `.claude/skills/skill-rules.yaml`. Triggers activate skills based on keywords, intent patterns, file paths, or content patterns, enabling progressive disclosure without manual invocation.

Claude Code skills support four trigger types—keywords, intent patterns, file paths, and content patterns—with three enforcement levels: block (prevents execution), suggest (recommends the skill), and warn (alerts users). Hooks like UserPromptSubmit and PreToolUse enable session tracking and conditional activation.

YAML frontmatter requires name and description fields at minimum. Place it at the top of SKILL.md, define trigger keywords and patterns, set enforcement level, and optionally specify hooks. This frontmatter configures which conditions activate your skill during Claude Code sessions.

The 500-line rule limits skill implementation scope to promote modularity and maintainability. Reference files provide detailed guidance without adding bulk, keeping core skill logic concise while supporting progressive disclosure of advanced features and configuration options.

Yes. Progressive disclosure uses reference files to surface essential information first while hiding detailed guidance. This keeps the main skill concise, follows the 500-line rule, and lets users access advanced patterns and examples through structured references without overwhelming the base skill file.

Test triggers using provided npx tsx commands to validate keyword matching, intent patterns, file path detection, and content pattern recognition. Refine patterns iteratively based on test results, then update SKILL.md frontmatter and skill-rules.yaml to deploy your trigger configuration.