Menu
refactor 流程的超轻量通道——直接识别 1-3 条低风险优化、一次确认、原地改、跑测试自证。触发：用户说"快速重构"、"小重构"、"简单优化下 XX 函数"、"别那么多步骤"，且改动在单函数 / 单组件局部、有测试可自证。
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
| name | cs-refactor-ff |
| description | refactor 流程的超轻量通道——直接识别 1-3 条低风险优化、一次确认、原地改、跑测试自证。触发：用户说"快速重构"、"小重构"、"简单优化下 XX 函数"、"别那么多步骤"，且改动在单函数 / 单组件局部、有测试可自证。 |
开始任何判断或动作前，先读取 .codestable/attention.md；缺失则视为骨架不完整，提示先补齐或运行 cs-onboard，不要回退到外部 AI 入口文件。
用户说"优化一下这个函数"而改动明显很小（单函数变长、组件里抽个 composable、一段重复代码合并）时走完整三阶段太重。fastforward 让 AI 像平时一样直接改但守住底线——行为等价、引用经典方法、跑测试自证。
很轻：没有 scan 清单、没有 design doc、没有 checklist，改完一句话汇报就行。
任一不过就退到 cs-refactor：
完整 scan 阶段是 7 条入场检查，这里压成最关键 3 条——剩下 4 条（跨模块 / 全口味 / 生成代码 / 扫不完）在"范围真的小吗"里已被隐含排除。
fastforward 不读完整方法库，但要守住"每一处改动都能对应到一个经典重构方法"。AI 心里想不出"我这步是 Extract Function / Memoization / Guard Clauses / ..." 里的哪一个 → 这次不是简单重构退完整流程查方法库。
常用方法（覆盖 fastforward 80% 场景）：
想做的动作不在这几种里、不是开箱即用的经典方法（涉及 Parallel Change / Strangler Fig / 分层纠偏）→ 退完整流程。
一句话回用户："我打算做 {方法名}，动 {具体文件/函数}，改动点 {N} 处，预计影响 {范围}。确认就开干。"
确认就下一步。用户说"还有个 X 要改"——评估 X 是否破坏入场 3 条，破坏了就退完整流程。
按经典方法步骤改。不产出 design doc / checklist，代码直接落盘。
✓ 已完成。方法：{方法名}。改动：{文件路径:行号范围}。验证：{跑了什么测试 / 通过情况}。
有偏离 / apply 过程中发现想再改点别的 → 停下问用户不发挥。
默认不建 .codestable/refactors/ 目录——fastforward 的价值就在不留存档。
例外：用户明确"这次要留个记录" → 建 .codestable/refactors/{YYYY-MM-DD}-{slug}/{slug}-refactor-note.md，内容就是上面那句汇报 + 一段"做了什么 / 为什么"。不写 design / checklist。
改到一半出现以下任一，停下告诉用户"比预期复杂，建议切回完整流程"：
切回：触发 cs-refactor 从 scan 开始。已改的部分要么提交保留、要么 git restore 回到干净状态再扫。
cs-refactor/SKILL.md — 完整 refactor 流程cs-refactor/reference/methods.md — 完整方法库.codestable/reference/system-overview.md — CodeStable 体系总览Iterate plans or implementations through fresh reviewer subagents until convergence. Use when the user explicitly asks for a "fresh reviewer loop", "plan iteration loop", "review and critique until clean", "spin up a reviewer to review my changes", or any phrasing where the goal is delegated adversarial review-then-fix cycles either on a plan before code or on a working implementation after code. Each round uses a new subagent so it has no prior-round context or groupthink.
通过企查查 OpenAPI 查询中国企业的工商、股东、司法、经营、知识产权、招投标、舆情等数据。当用户说"查 XX 公司"、"看 XX 的工商/股东/有没有官司/有没有被执行/有没有商标专利/有没有经营异常"、"XX 是不是空壳"、"XX 法人是谁"、"XX 注册资本"、"查企查查"、"查企业"等任何企业信息查询需求时，使用此技能。已注册 45 个商业接口（用户需自行在企查查后台逐个申请开通）。零依赖纯 Python，跨平台 Mac/Windows/Linux。
Use this skill when the user is working with LyteNyte Grid (@1771technologies/lytenyte-pro or @1771technologies/lytenyte-core), a headless React data grid. Activate for tasks like: installing or licensing the grid, configuring columns or rows, building cell renderers or editors, adding filters or sort controls, grouping or aggregating rows, pivoting, exporting to CSV/Excel/Parquet/Arrow, row selection, cell range selection, theming or styling, TypeScript GridSpec patterns, server-side or tree data, and any PRO component (SmartSelect, PillManager, Menu, Dialog, TreeView, RowGroupCell). Also activate when the user describes grid problems without naming the package — e.g. "my rows won't group", "cells aren't editable", "add a loading overlay", "pin this column", "the filter isn't working", "how do I export this table", "select a range of cells", "copy cells to clipboard".
Design websites and applications that AI agents can consume, navigate, and interact with. Use when building any site, app, or product that agents will use as an end-user — not just crawl or index. Covers semantic structure, accessibility-as-agent-interface, machine-readable data, API-first patterns, and the emerging protocols (llms.txt, MCP, NLWeb, A2UI) that make sites agent-ready. Triggers on: agent-friendly, agent-readable, agent-accessible, AX, agent experience, agentic web, dual-interface, machine-readable, llms.txt, MCP integration, NLWeb, accessibility tree, ARIA for agents, structured data, JSON-LD, Schema.org, API-first design, build for agents, agent-ready.
Design system for AI agents that build UI. Automatically routes to the right quality checks based on the task. Triggers on ANY visual, frontend, UI, design, component, page, layout, or styling work. Includes: anti-pattern detection, state completeness checks, accessibility verification, typography/color/spacing guidance, and creative direction when needed. Install this one skill to get the full system — it orchestrates everything else.
Core pack — always active for visual work. Quality gate for UI, components, pages, layouts, or frontend work. Triggers on any visual/design task automatically. Use before presenting work, during builds, and for design QA.