Skill Authoring
Start with the observed problem and the professional judgment the skill adds. Write a succinct description that lets the model recognize when that judgment helps. Avoid triggers that make every task load the skill.
Give the reader the distinctive questions, distinctions, examples, or tool pitfalls needed to work well. State necessary ownership, permission, and side effect boundaries precisely. Put deeper specialist material in references when it improves readability.
Let the model choose methods and sequence. Do not recreate task tiers, compulsory step chains, arbitrary worker or review counts, stop-status taxonomies, mandatory reports, anti-rationalization sections, or a fixed skill anatomy. Do not repeat Core principles in each skill or implement personalization inside a skill.
For delegation guidance, preserve independent contexts, evidence gathering, and conclusions. Separate source facts and settled decisions from parent hypotheses; name write scopes and let the parent integrate and verify.
Check that the skill's description, links, examples, and shipped copies match its actual purpose. Validate relevant behavior and build consistency, not a ritual heading structure or a target number of skills.