cs-run

name	cs-run
description	Use as the main CS Skills entrypoint when the user does not explicitly name a downstream skill, asks Codex to execute or organize a task, or the goal is still unclear. Trigger for "$cs-run", "cs-run", "帮我执行", "帮我处理", "不知道用哪个 skill", "先规划再做", and vague product, content, engineering, design, or workflow requests. Turn intent into a compact Goal Card, ask only blocking questions, route to exactly one active skill, and continue through that skill when execution is requested.
metadata	{"author":"陈硕","collection":"CS Skills","source":"https://github.com/ChenShuo2004/cs-skills","compatibility":"Codex and any agent that supports SKILL.md"}

CS Run

Purpose

cs-run is the single front door for this skill library. It converts a rough request into a clear Goal Card, selects one active downstream skill, and keeps the task moving instead of making the user learn the directory structure.

Use it when the user says what they want to achieve but does not know which skill to use. If the user explicitly names another active skill and provides enough input, follow that skill directly.

Active Routes

Choose one primary route from this table:

User goal	Route	Typical output
Write, outline, rewrite, polish, or turn project material into an article	$cs-writer	Angle, outline, draft, rewrite, or review
Research a product, company, technology, market, or competitor set for a decision	$cs-search-skill	Source-backed decision brief, comparison, risks, and next actions
Design, build, revise, or review a user-facing web interface	$cs-frontend-design	UI plan, implementation guidance, and browser checks
Clean up code, reconcile implementation with requirements, or prepare a maintainable handoff	$cs-clean-code	Scoped edits, documentation sync, tests, and verification
Save a rollback point before risky repository changes	$cs-checkpoint-version	Restorable local checkpoint and verification
Turn a Markdown PRD into a Ralph workflow or dry-run	$cs-ralph-runner	Ralph PRD, overview, dry-run output, and logs
Verify, commit, push, open a PR, or deploy a completed change	$cs-ending-time	Delivery verification and GitHub/Vercel handoff
Generate, edit, convert, or extend the fixed 小黄 / “有温度” brand IP while preserving character identity	$cs-xiaohuang-skill	2D/3D character assets, action variants, collaboration art, identity QA, or article illustration shot lists
Replicate an ecommerce short video or create storyboard/Seedance/Flow/Veo packages	$cs-auto-videl	Storyboards, prompts, generation package, and QC
将单个或一批内容想法收敛为 ChatCut 短视频，或了解从策划到成片的上手流程	$cs-chatcut	选题排序、中文口播稿、素材清单、Motion Graphics、声音方向、逐镜头表和操作指南
The user has not described a usable goal yet	$cs-run	Goal Card and the smallest useful questions

Retired Routes

Do not route to removed skills:

• Automatic editing, timeline rendering, MP4 export, and automatic-editing Ralph workflows are retired from this library.
• Li Auto infographic workflows are retired from this library.
• Open Design artifact workflows are retired from this library.

If a request belongs to a retired route, say that the route is no longer active and ask whether the user wants a new dedicated skill. Do not silently send a generic editing request to $cs-auto-videl; it is for ecommerce creative and generation workflows.

Workflow

1. Inspect the current repository and nearby project documentation before asking questions.

2. Restate the intended result in one sentence.

3. Fill this Goal Card:

   Goal:
   Inputs:
   Expected output:
   Audience/user:
   Constraints:
   Validation:
   Recommended skill:
   

4. Ask only questions that can change the route or make execution unsafe. Ask up to three concise blocking questions; record safe defaults instead of asking about preferences that do not matter yet.

5. Select exactly one primary skill. Add a secondary skill only when it is a necessary verification or delivery step.

6. If the user asked to execute, continue through the recommended skill in the same task. If the user only asked which skill to use, return the Goal Card and a ready-to-send prompt.

7. Verify the result against the requested output and validation criteria before reporting completion.

Routing Rules

• Prefer the narrowest active skill that matches the requested outcome.
• Do not route by a single keyword when the user's actual output is clear.
• Respect explicit skill names. $cs-writer, $cs-search-skill, $cs-frontend-design, $cs-clean-code, $cs-checkpoint-version, $cs-ralph-runner, $cs-ending-time, $cs-auto-videl, $cs-xiaohuang-skill, and $cs-chatcut take precedence over generic routing.
• Route requests about “小黄”、“有温度 IP”、“温度种子”、空心爱心天线、固定角色延展、2D/3D 转换、联名、风格迁移或身份修复 to $cs-xiaohuang-skill. Do not use it for unrelated general image generation.
• Route requests containing 调研、竞品分析、深度研究、市场判断 or similar intent to $cs-search-skill when the user needs evidence for a decision. Do not use it for a simple definition or one-fact lookup.
• 将中文文章、观点或方法论转成正文配图的请求也路由到 $cs-xiaohuang-skill；将内容想法、短视频主题或 ChatCut 上手请求路由到 $cs-chatcut；文章或长文改写交给 $cs-writer，实际 ChatCut 项目编辑和电商视频复刻继续使用各自的专用工作流。
• Do not ask the user to choose a skill when the route is unambiguous.
• Do not invent a missing domain skill. Report the retired or unsupported route and propose the smallest next decision.
• Keep the Goal Card short. It is an execution aid, not a long questionnaire.

Output Contract

When routing only, report:

我理解的目标：
推荐 skill：$<skill-name>
理由：
默认假设：
需要补充的信息：
下一步：

When executing, route immediately and report the downstream skill's actual result, changed files, validation, and remaining risk.