agent-skill-designlisted

# Agent Skill Design ## Purpose Design Agent Skills as selectable behavior modules for recurring task classes. Focus on what the agent should do differently, when that behavior should apply, where its responsibility ends, and how to tell whether it helped. Defer packaging, file layout, metadata syntax, templates, installation, and platform-specific validation to the applicable authoring skill or specification. ## Core Criterion Create or retain a Skill only when it improves observable agent behavior for a recognizable, recurring task class. Do not use a Skill merely to preserve information, repeat universal rules, or restate behavior the model already performs reliably. Require a concrete answer to all of these questions: - What recurring task does the Skill improve? - What observed or credible failure does it prevent? - What should the agent do differently because the Skill exists? - When should and should not that behavior activate? - What evidence would show an improvement over the current baseline? If the behavioral difference or evidence cannot be stated, gather examples or keep the content outside a Skill until the need is clearer. ## Design Procedure ### 1. Establish Evidence And Baseline Start from representative tasks, artifacts, feedback, traces, or observed failures. Distinguish observed problems from hypotheses. Record: - representative requests that expose the need - how the agent behaves without the Skill - the failure, waste, risk, or inconsistency