MiniAiWorkflow

Use when this project needs Docs initialization, Docs structure maintenance, or Obsidian-friendly documentation without leaving the project's Docs system.

Use when an M or L level task in this project needs design clarification, option comparison, or user-approved solution shaping before planning or execution.

Use when a bug, test failure, compile failure, or unexpected behavior in this project needs root-cause analysis before any fix is attempted.

Use when a task in this project needs a clearer proposal or implementation plan while keeping all documents inside the project's Docs system.

Use when this project needs code review handling or structured feedback processing without turning every small task into a heavy mandatory gate.

Use when working inside this project and a stronger workflow layer is needed for planning, debugging, verification, or task decomposition without leaving the project's Docs system.

Use when about to say work is complete, fixed, or ready in this project and fresh verification evidence is required first.

L 级任务前期或需跨多份文档/模块时，并行 spawn 只读子智能体分别分析切片后由主智能体统一汇总。适合多文档分析、跨模块摸底、风险盘点。

M/L 任务中，针对 >=3 个互不依赖的模块并行 spawn 可写子智能体。主智能体先定义接口契约与文件分配表，再做集成审查与验证。

项目主代码目录为空或极少且用户确认启动正式开发时，并行 spawn 3 个子智能体：项目骨架 + 核心渲染 + 核心玩法。根据实际引擎类型（Unity/Godot/Unreal/Web/其他）和项目特点自适应调整任务内容。

实现完成后可运行完整 team-review；跨模块计划成稿前可单独复用 architecture-review 角度做预审。支持并行（Claude Code）和降级顺序（OpenCode）双模式。

在 AI 协作启动工作流初始化阶段，自动创建 .claude/ 配置（路径规则 + hooks + settings.json）。用户在初始化对话中确认启用后，AI 读取本技能并按模板生成所有文件。

专项约束接线技能。把用户提供的行业规范、项目命名约定、模块规则或技能草稿注册为工作流可加载的专项约束包。

招聘与定制子智能体。根据项目需求推荐 agent 组合，生成带有个性化设定的子智能体定义文件。不执行代码，只产出配置。