Matt Pocock describes the problem of assembling agent skillsAn AI agent skill is a reusable package of instructions, resources, and tool guidance for performing a bounded kind of work. without a way to evaluate themEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria.. He proposes examining a skill through four dimensions: its trigger, structure, steering and pruning.
Matt Pocock compares user-invoked and model-invoked skills, explaining their respective cognitive load and context costs. He separates procedural steps from reference material and suggests moving branch-specific material into separate files.
Matt Pocock explains how concise leading words and focused phases can guide agentsPrompt engineering is the practice of designing and refining instructions, context, examples, and constraints to obtain useful AI outputs., then discusses pruning duplicate, stale and ineffective instructions. He presents his checklist as a practical way to review and improve skills.
Watch on YouTube
