skill-developer

Creating a Claude Code skill requires proper YAML frontmatter defining metadata, triggers, and enforcement levels. The skill body contains instructions and logic, adhering to Anthropic's best practices for robust AI assistance and clear trigger mechanisms.

Claude Code skills support trigger types including keywords, intent patterns, file paths, and content patterns. Enforcement levels dictate behavior through block, suggest, or warn mechanisms, allowing precise control over when and how a skill activates within the AI assistant workflow.

If your Claude Code skill is not activating or has false positives, check the trigger conditions in your skill-rules.json. Debugging involves verifying hook mechanisms like UserPromptSubmit and PreToolUse, ensuring intent patterns are specific, and confirming the 500-line rule is not exceeded.

Hook mechanisms like UserPromptSubmit and PreToolUse intercept user prompts and tool executions to enforce skill rules. UserPromptSubmit triggers when a user submits a prompt, while PreToolUse hooks run before a tool executes, allowing skills to block, warn, or suggest actions dynamically.

The 500-line rule is a best practice constraint in skill development limiting skill complexity. Adhering to this rule ensures the skill remains maintainable, performs efficiently, and avoids overwhelming the AI assistant context, which helps prevent activation issues and debugging difficulties.

Yes, you can use session tracking to manage Claude Code skill enforcement. It allows skills to maintain state and context across interactions, enabling more sophisticated enforcement levels like block, suggest, or warn based on user behavior and previous prompt submissions within the session.