skill-developer

Creating a Claude Code skill involves defining file structure, configuring triggers using keyword and intent patterns, setting enforcement rules, and implementing hooks like UserPromptSubmit and PreToolUse. Start with the skill creation workflow guide, then test your triggers against representative code to ensure they activate correctly.

Guardrail skills block actions by enforcing rules and preventing errors, while suggestion skills provide proactive guidance. Guardrails use block enforcement; suggestions use PreToolUse hooks to offer recommendations. Choose guardrails when you need to prevent mistakes, suggestions when you want to guide without restricting.

Skill triggers activate based on keyword patterns, intent detection, file paths, and content matching. Define patterns declaratively in your skill's trigger configuration, test them against multiple code contexts, and use progressive disclosure to surface guardrails only when relevant to avoid overwhelming users.

Skill triggering failures usually stem from overly broad or narrow trigger patterns, misaligned intent detection, or content matching that doesn't capture your target scenario. Debug by testing each trigger component independently, review your pattern definitions against real code samples, and verify your skill respects the 500-line limit.

Yes. The UserPromptSubmit and PreToolUse hooks enable real-time suggestions by intercepting user input and tool calls. Use PreToolUse to analyze planned actions and offer guidance before execution, or UserPromptSubmit to detect intent and suggest best practices proactively.

Follow Anthropic's guidelines including the 500-line rule for skill size, use deterministic trigger patterns, implement session tracking for enforcement, test across diverse code contexts, and version your skill metadata clearly. Progressive disclosure prevents overwhelming users with simultaneous guardrails.