cli-forge
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
Chinesecli-forge
cli-forge
Use this parent skill when you want one stable entrypoint for the full
workflow instead of naming a stage skill up front.
cli-forgeThis skill does not replace the child skills. Its job is to identify the
current stage, recover the earliest incomplete phase from repo state, and then
route work to the matching child skill.
当你需要为整个工作流提供一个稳定的入口,而不需要提前指定阶段skill时,请使用这个父skill。
cli-forge本skill不会替代子skill。它的职责是识别当前所处的阶段,从仓库状态中恢复最早的未完成阶段,然后将工作路由到对应的子skill。
Canonical References
规范参考
../cli-forge-intake/SKILL.md../cli-forge-description/SKILL.md../cli-forge-scaffold/SKILL.md../cli-forge-extend/SKILL.md../cli-forge-validate/SKILL.md../cli-forge-publish/SKILL.md
Read this file first, then move into the child skill that owns the current
phase.
../cli-forge-intake/SKILL.md../cli-forge-description/SKILL.md../cli-forge-scaffold/SKILL.md../cli-forge-extend/SKILL.md../cli-forge-validate/SKILL.md../cli-forge-publish/SKILL.md
请先阅读本文件,然后进入负责当前阶段的子skill。
Parent Role
父角色职责
The parent skill is responsible for:
- classifying the request as ,
intake,description,scaffold,extend, orvalidatepublish - inspecting the current filesystem state when the request is ambiguous
- selecting the earliest incomplete stage instead of skipping ahead
- routing into the child skill that should do the actual work
- supporting resume flows when the user says only "continue cli-forge" or provides partial context
The parent skill should stay thin. It should not duplicate the detailed
instructions already owned by the child skills.
父skill负责:
- 将请求分类为、
intake、description、scaffold、extend或validatepublish - 当请求含义模糊时,检查当前文件系统状态
- 选择最早的未完成阶段,而不是跳过前置步骤
- 将工作路由到应该执行实际任务的子skill
- 当用户只说“继续cli-forge”或只提供部分上下文时,支持恢复执行流程
父skill应当保持轻量,不应重复子skill中已经存在的详细说明。
Stage Selection
阶段选择规则
Choose the child skill using the request plus the current project state:
- Route to when the request is ambiguous, when required inputs are missing, or when the correct stage is not obvious yet.
intake - Route to when the generated skill's purpose, positioning, or user-facing contract must be created or refreshed before implementation.
description - Route to when the user wants a brand-new Rust CLI Skill project and the approved description contract already exists.
scaffold - Route to when a scaffolded project already exists and the requested change is exactly
extendorstream, or when description-impacting work must continue into feature changes.repl - Route to when the user explicitly wants an audit or when freshly scaffolded or extended work needs verification.
validate - Route to when validation is current and the workflow needs final closure, dry-run review, rehearsal, destination-config checks, or the documented release path.
publish
If more than one stage seems plausible, inspect the filesystem and route to the
earliest stage that is not clearly complete.
结合请求内容和当前项目状态选择子skill:
- 当请求含义模糊、缺少必填输入,或是暂无法明确正确阶段时,路由到
intake - 当需要在实现前创建或更新生成skill的用途、定位或面向用户的约定时,路由到
description - 当用户需要一个全新的Rust CLI Skill项目,且已存在获批的描述约定时,路由到
scaffold - 当已存在初始化好的脚手架项目,且请求的变更恰好是或
stream,或是涉及描述修改的工作需要继续推进功能变更时,路由到replextend - 当用户明确要求审计,或是刚刚完成脚手架搭建或扩展的工作需要验证时,路由到
validate - 当验证已完成,工作流需要最终收尾、dry-run审核、演练、目标配置检查,或是执行已文档化的发布路径时,路由到
publish
如果看起来有多个阶段都适用,请检查文件系统,路由到最早的未明确完成的阶段。
Resume Rules
恢复执行规则
- If no target project directory exists yet, resume at , then continue into
description.scaffold - If the directory exists but is missing the scaffold baseline, route back to
rather than forcing feature or validation work through.
scaffold - If the project exists and the request is to add or
stream, resume atrepl.extend - If scaffold or extend work just completed, always continue into before treating the workflow as done.
validate - If the user wants release work but validation is stale or missing, route back
to first.
validate - If validation is current and the user did not explicitly ask for a release
side effect, still continue into so the flow ends with
publishclosure instead of stopping half-finished.report_only
- 如果还不存在目标项目目录,从阶段恢复,然后继续进入
description阶段scaffold - 如果目录存在但缺少脚手架基准文件,回到阶段,而不是强制推进功能或验证工作
scaffold - 如果项目已存在,且请求是添加或
stream,从repl阶段恢复extend - 如果脚手架或扩展工作刚刚完成,在认定工作流完成前,始终先进入阶段
validate - 如果用户需要执行发布工作,但验证已过期或缺失,先回到阶段
validate - 如果验证已完成,且用户没有明确要求执行发布操作,仍需继续进入阶段,以便流程以
publish模式收尾,而不是中途停止report_only
Typical Flows
典型流程
- New project: ->
intake->description->scaffold->validatepublish - Add feature: ->
intake->extend->validatepublish - Description update: ->
intake->description->extend->validatepublish - Validation request: ->
intake->validatepublish - Release-oriented request: ->
intake->validate, unless validation is already currentpublish
- 新项目:->
intake->description->scaffold->validatepublish - 添加功能:->
intake->extend->validatepublish - 更新描述:->
intake->description->extend->validatepublish - 验证请求:->
intake->validatepublish - 发布相关请求:->
intake->validate,除非验证已经是最新状态publish
Routing Examples
路由示例
- "Create a new cli-forge skill" -> start with , then
intake.description - "Add repl to this generated skill" -> inspect the target project, then route
to if the scaffold baseline is present.
extend - "Check whether this skill is compliant" -> route to .
validate - "Do a release dry run" -> route to only after confirming validation is current.
publish - "Continue cli-forge" -> inspect the repo and resume from the earliest incomplete stage.
- "创建一个新的cli-forge skill" -> 从开始,然后到
intakedescription - "为这个生成的skill添加repl功能" -> 检查目标项目,如果存在脚手架基准文件,就路由到
extend - "检查这个skill是否符合规范" -> 路由到
validate - "执行发布dry run" -> 仅在确认验证已完成后,路由到
publish - "继续cli-forge流程" -> 检查仓库,从最早的未完成阶段恢复执行
Done Condition
完成条件
This parent skill is complete for the current request only when:
- the correct child skill has been selected
- the required inputs for that child skill are assembled or recoverable
- the handoff is explicit enough that work can continue without re-triage
仅当满足以下条件时,本父skill对当前请求的处理才算完成:
- 已经选好了正确的子skill
- 该子skill所需的输入已经收集完毕或可恢复
- 交接足够清晰,不需要再次分类即可继续推进工作