designing-workflow-skills

Best practices for structuring Claude skills involve dividing logic across SKILL.md, references, and workflows. This ensures robust execution by using progressive disclosure, appropriate tool assignment, and avoiding common anti-patterns.

Choosing the right workflow pattern for LLM architecture depends on task logic: use Sequential Pipeline for ordered steps, Routing for decision points, Linear Progression for simple flows, Safety Gate for validation, and Task-Driven for user interaction.

You need to refactor existing code structure for Claude Code skills when execution becomes unreliable or complex. Refactoring applies structural anatomy principles and anti-pattern fixes to improve maintainability and reliable task execution.

Progressive disclosure in skill design is the practice of splitting content across SKILL.md, references, and workflows. It prevents overwhelming the LLM by only exposing necessary instructions and tool assignments at each specific execution step.

Sequential pipeline patterns handle ordered, multi-step execution without branching. For tasks requiring decision points, routing patterns are recommended to properly direct the workflow logic based on specific conditions.

Claude skill execution fails reliably due to common anti-patterns like poor tool assignment or overloaded SKILL.md files. Fixing these structural issues through proper workflow patterns and progressive disclosure ensures robust performance.