菜单
When the user asks to refactor, restructure, extract, rename, simplify, or clean up code without changing behavior, consult this skill to decide whether to delegate to the refactor subagent and how to craft the prompt.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Record a durable engineering decision in the repo's append-only decision log (docs/decisions.md by default). Use after choosing between real alternatives — a dependency, a version pin, an architecture or protocol choice, a deliberate non-implementation, or a reversal of an earlier decision. Also use when the user says "log that decision", "write this down", or asks what was decided and why.
Structural code search and refactoring via ast-grep. Reach for this when Edit fails with "old_string not unique", when refactoring across formatting variations, or when matching code shape rather than text.
Outline a file or package (functions, classes, methods, types, imports, exports) instead of reading it whole. Use before opening an unfamiliar or large file to orient, when surveying a package's API surface, or to find where a symbol is defined so you can Read just that range -- any "what's in / list / show me the structure of" need. Not for "where is X used" (search) or "how does X work" (read the body).
How to author Claude Code plugins (agents, skills, commands, hooks, MCP servers) and distribute them via a marketplace. Use when creating, structuring, or revising a plugin, deciding which skill pattern fits, or writing SKILL.md / agent / hook / manifest files.
Implement TODO.md items via isolated subagents, then adversarially verify correctness. Drives a phased loop that implements one checkbox at a time, runs independent lint/test/review subagents, handles human verification, and commits per subsection.
Create and maintain structured TODO.md implementation plans with phased tasks, testing philosophy, and verification checklists. This skill should be used when the user asks to create a TODO file, implementation plan, project roadmap, or phased development plan. Also use when asked to update or extend an existing TODO.md in this style.
| name | how-to-refactor |
| description | When the user asks to refactor, restructure, extract, rename, simplify, or clean up code without changing behavior, consult this skill to decide whether to delegate to the refactor subagent and how to craft the prompt. |
| user-invocable | false |
Delegate to the refactor subagent (Agent tool, subagent_type: refactor) when the user wants code restructured without behavior change. The subagent executes one atomic refactoring, runs tests to verify behavior preservation, and stops. Doing this work in the main session risks tangling refactor edits with unrelated changes and consumes context on diff-level reasoning.
For broader "what should I refactor?" questions, use plan-refactor instead — see how-to-plan-refactor.
Delegate when:
plan-refactor report and asks you to carry it out.Handle inline (do not delegate) when:
The subagent requires a specific target. Pass exact instructions: what to refactor, what technique to apply (if known), what scope (one site vs. all instances).
Examples:
getUserData to fetchUserData across the codebase. Verify tests pass."plan-refactor to identify one).