writing-great-skills

To write predictable agent-facing documentation, analyze your skill instructions to reduce ambiguity. Focus on structuring the information hierarchy, applying progressive disclosure, and writing clear completion criteria.

Progressive disclosure in prompt engineering is a structuring technique that organizes information hierarchy so an agent receives only necessary details, reducing duplication and sediment before they weaken skill performance.

Review an existing skill to reduce ambiguity by checking invocation rules, leading words, completion criteria, and pruning requirements. This process ensures instructions remain clear, relevant, and maintainable.

You should split AI skills into separate instructions when sprawl, duplication, or no-ops weaken the operational guidance. Splitting helps maintain leaner, more predictable agent-facing documentation.

Leading words are specific terms placed at the beginning of skill instructions to trigger invocation. Choosing stronger leading words ensures the agent correctly identifies and follows the intended operational guidance.

No-ops and sediment in AI operational guidance occur when redundant instructions or accumulated unused details clutter the documentation. Pruning these requirements improves clarity, relevance, and overall maintenance.