refactor-skill

Refactoring LLM resources and prompts is the intentional restructuring of prompt content, tool definitions, and supporting context (e.g., skills, instructions, examples, guardrails) to improve composability, clarity, reuse, and invocation precision without changing the underlying capabilities, knowledge coverage, or output semantics of the original monolithic prompt.

At the architecture level, this refactoring decomposes a monolithic LLM skill into smaller, purpose-built, independently invocable skills with well-defined responsibilities, inputs, and outputs—reducing coupling and cognitive load, enabling targeted reuse and testing, and preserving behavioral parity through explicit contracts, shared primitives, and regression validation against the original monolith.

Refactoring Workflow

You are focused on refactoring the skill found at the path: $ARGUMENTS

If no path is provided STOP and say: "No path provided. Please provide a path to a skill or plugin to refactor." If the path points to a claude code plugin load the Skill claude-plugins-reference-2026 which provides the current spec and details for how a plugin is defined, when provided a plugin path, you will be refactoring the skills found within the plugin skills/ directory.

The paths provided below, like references/ are relative to the skill directory.

Phase 0: Candidate Assessment (GATE)

Before analyzing for a split, determine whether skill splitting is the appropriate action at all.