refactor - SKILL.md Agent Skill

name: refactor description: Produce a structured refactor plan for a file or directory against this project's architecture and testing rules. Read-only — outputs a plan, does not edit. Use when the user asks to plan a refactor, audit a file's structure, or check if a hook/component/resolver needs splitting. disable-model-invocation: true allowed-tools: Task

Invoke the refactor-planner subagent to plan the refactor.

Scope:

$ARGUMENTS must be a path (file or directory) inside apps/. Pass it verbatim to the agent as its first user message.
If $ARGUMENTS is empty, ask the user which file or directory to plan before invoking the agent.
Extra text after the path (e.g. "focus on the analytics") is also passed verbatim as scope guidance.

Return the agent's plan as-is. The output format (PLAN: line + Violations / Splits / Extractions / Test plan / Order of operations / Risks) is designed to be read directly — do not re-summarise or filter.

If the plan ends with clean, a one-sentence acknowledgement is fine. Otherwise surface the plan verbatim so the user can approve it or tweak it before any edits happen.

The agent never edits files. After the user approves the plan, applying it is a separate step (Claude Code in normal mode, with the plan as context).