writing-skills

To standardize agent skill creation, use decision trees, ready templates, and authoring standards that enforce consistent SKILL.md frontmatter, discovery triggers, and anti-rationalization discipline across all deployed skills.

The best way to structure a sprawling SKILL.md is to modularize it using a decision tree dispatcher to select Tier 1-3 architectures, then apply technique, reference, pattern, or platform templates for scalable progressive disclosure.

Agents ignore custom skills due to weak discovery triggers, overlooked anti-rationalization controls, or missing tests. Fix this by applying discovery-optimized description guidance and enforcing RED-GREEN-REFACTOR testing checklists.

Choose the right architecture tier by using a decision tree dispatcher that evaluates skill complexity and guides you to Tier 1, Tier 2, or Tier 3 architectures for appropriate technique, reference, and platform implementations.

To verify metadata and enforce anti-rationalization, apply CSO checklists, review rationalization red flag tables, and execute RED-GREEN-REFACTOR testing checklists before deployment to ensure strict agent compliance.

Yes, you can fix broken discovery triggers for existing agent skills by validating naming conventions, optimizing YAML frontmatter descriptions, and applying discovery-optimized standards to ensure agents find and follow the skills consistently.