skill-developer

Create a Claude Code skill by defining SKILL.md frontmatter, configuring skill-rules.json with trigger types (keywords, intent patterns, file and content patterns), and setting enforcement levels (block, suggest, warn). Use hook mechanisms like UserPromptSubmit and PreToolUse to activate guardrails that prevent unsafe edits and misconfigurations across your automation workflows.

Follow best practices including the 500-line rule for skill complexity, progressive disclosure to reveal capabilities incrementally, and session tracking to maintain context. Define clear trigger conditions, implement appropriate guardrails for your domain, and use blockMessage templates with environment overrides so operators can adjust enforcement dynamically without code changes.

Skill activation uses two hook types: UserPromptSubmit fires before user input reaches Claude, and PreToolUse fires before tool execution. Triggers include keywords, intent patterns, file path matching, and content patterns. Together, they determine when a skill activates and enforce domain-specific rules across code edits, file opens, and tool invocations.

Yes, guardrail skills block unsafe operations or suggest best practices using enforcement levels. Define rules in skill-rules.json, set conditions under which they apply, and use skipConditions to exclude specific cases. Environment overrides let operators control enforcement without rebuilding, making guardrails adaptable to different projects and risk profiles.

Build skills using npm and TypeScript. You'll define SKILL.md frontmatter for metadata, create skill-rules.json following the predefined schema, and configure trigger patterns and enforcement policies. References components provide lookup data; session tracking and progressive disclosure are built-in mechanisms you configure rather than implement from scratch.