starworkInit

starworkInit

Use this skill to turn the user's request to "create / connect / initialize a StarWork workspace or project hub" into a safe process of checking, previewing, confirming, and writing.

starworkInit is not starwork init the command itself. The Skill is responsible for judgment, explanation, interviewing, previewing, confirming, and integrating entry rules; the CLI is responsible for actually generating the dry-run, writing workspace files, and running the doctor check.

Main entry boundary

If the user is only asking about the product overview, getting-started path, installation entry, or which StarWork capability to use, go back to the starwork main entry.starworkInit only handles requests that already point to creating, connecting, or initializing a project workspace or project hub.

Unless the user is still in the discussion stage, do not stop at a blueprint or suggestion. When the user explicitly says "create, initialize, generate, or transform into a workspace," you must continue running the CLI: first starwork init --dry-run, then after getting user confirmation starwork init --yes, and finally run or suggest running starwork doctor --target <path> to verify.

Before executing --yes , you must first have the user confirm the final absolute path. The AI may suggest a folder name, but must state that it can be modified; do not treat the suggested folder name as a path the user has already agreed to.

Reference loading rules

Before hitting a specific scenario, first read the corresponding reference. If the reference file does not exist or cannot be read, high-risk actions must stop, prompt the user with "The Skill installation is incomplete, please reinstall StarWork Skills using the complete directory," and state the missing file path. A complete installation means the Skill main file,references/ , and agents/ are installed together with the directory.

High-risk actions include: executing --yes, merging entry rules, modifying host rules, generating an executable blueprint, and writing custom rules.

Scenario	Required reference
Determining user intent	references/intent-routing.md
First-time initialization or MultiAgent reflow	references/friendly-onboarding.md, references/initialization-flow.md
Standard project / hub initialization	references/initialization-flow.md, references/output-and-safety-rules.md
Custom directory or rules	references/custom-blueprint.md, references/output-and-safety-rules.md
Existing non-empty project	references/existing-project-agent-docs.md, references/output-and-safety-rules.md
Host adaptation	references/host-adapter-flow.md, references/existing-project-agent-docs.md
User asks about knowledge base / Pack	references/knowledge-and-pack-boundaries.md

references/README.md explains complete directory installation and the rule to stop when a reference is missing. Do not pile long processes, complete JSON blueprint examples, or large command sequences back into the main Skill.

The first screen must explain the product first

When the user asks for the first time to initialize, create, connect, or transform into a StarWork workspace, do not immediately ask for configuration or give commands directly. First explain in the user's language:

可以，我先简单说清楚 StarWork 在做什么。

StarWork 是给 AI 协作准备的项目工作台。它会把项目说明、当前任务、协作规则、交接记录和健康检查入口放到固定位置，让 AI 每次进入项目时不用从零猜上下文。

这次我会带你做三件事：
1. 确认这个工作台服务哪个项目；
2. 预览 StarWork 准备补哪些协作文件；
3. 你确认后再正式写入，并做一次检查。

我会先预览，不会直接改你的业务代码，也不会直接覆盖已有 AI 规则文件。

After explaining the product, then move to the first question, and explain why you are asking. Ask only one question at a time. Do not ask about type, language, path, structure, knowledge base, and host all at once.

MultiAgent-only reflow first screen

When the user is reflowing from starworkMultiagent , or the user only says "I want to enable multi-AI collaboration / multi-Agent division of labor, but this project has not been connected yet," start with the following user-language passage. Do not first talk about commands, parameters, or internal mechanisms:

我先把你带回安全接入这一步。

不是要打断你创建多 AI 分工，而是多个 AI 开始协作前，需要先确认三件事：
1. 这个项目是什么；
2. 当前正在推进什么；
3. 不同 AI 哪些内容能整理，哪些内容不能随便改。

我会先做一份写入预览，只告诉你准备补哪些协作入口和边界。
确认前不会改业务内容，也不会覆盖已有 AI 规则文件。

Then ask only one question:

你现在是想把当前已有项目接入 StarWork，还是新建一个空项目工作台试用？

MultiAgent quick-start rules

When the user's goal is "enable multi-AI collaboration first," by default ask only for the minimum necessary information:

1. Target directory.
2. New or existing project.
3. Workspace language.
4. Main AI tool used; can be skipped if unsure.
5. If it is an existing project, read the existing entry and generate a draft to be integrated.

Do not expand the knowledge base, Packs, Capabilities, complex directory customization, or multi-host adaptation by default. Only enter the full initialization interview when the user actively requests customization.

Before each question, explain the purpose first: an existing project needs its original files protected; the target directory determines the preview; the language affects directory names and rule expression; the primary AI tool determines the entry file, and not all tool rules will be changed at once.

Check / Preview / Write layering

Before any write, you must check and preview first:

• The check phase is read-only.
• dry-run is a preview and does not write files.
• --yes can only be executed after the user confirms the absolute path, write scope, and risks.
• After completion, run or suggest running starwork doctor --target <path>。

The dry-run recap must include:

这是预览，还没有写入文件。

准备新增：
- ...

准备更新：
- ...

不会改动：
- 你的业务代码
- 已有非空 AI 规则文件

需要你确认：
- 目标路径是否正确
- 是否接受这些 StarWork 协作入口
- 是否继续合并待整合草稿

Existing projects and pending merge

Before a dry-run on an existing project, you must first explain:

我检测到这是已有项目时，会按更保守的方式处理：
- 保留你的业务代码和现有资料；
- 不直接覆盖已有 AGENTS.md、README.md、CLAUDE.md 或其他 AI 规则入口；
- 先生成待整合草稿；
- 让你确认后再合并入口说明。

An existing non-empty project must go through --agent-docs draft. Do not directly overwrite AGENTS.md、README.md、CLAUDE.md or the host entry; after reading .starwork/drafts/agent-docs-plan.json and the proposed draft, integrate according to the project context.

If pending merge / pending_merge / pending integration status appears, do not say that multi-AI collaboration is fully available. Explain in layers:

工作台骨架已经写入，但 AI 入口还没有最终生效。
因为你的项目已有规则文件，我只生成了待整合草稿，没有直接覆盖。

下一步我会把现有规则和 StarWork 建议放在一起，让你确认后再合并。

Success status wording

Success reporting must be layered; do not mix it into a single status:

• Check completed.
• Write preview completed.
• Workbench skeleton has been written.
• AI entry is pending integration or already in effect.
• The doctor check passed or there are still pending items.
• MultiAgent can continue designing AI roles, or the entry / boundaries need to be completed first.

Safety boundaries

• Do not overwrite the user's existing business files.
• Do not write the initialization report, dry-run results, doctor results, or npm installation process into AGENTS.md or .starwork/rules/*.md。
• Do not explain internal terms such as adapter profile、host_native_dirs、workspace schema to novice users.
• Do not let the blueprint determine the real target path; the target path must come from command parameters or explicit user confirmation.
• Do not put the knowledge base and Packs into the default interview flow; only read the corresponding reference when the user actively requests it.