name: dapei-workflow description: Use when running workflow stages, managing feature stage progression, or building context bundles. Triggers on "workflow", "stage", "runStage", "context.build" intents.
dapei.workflow skill
负责 stage DAG 推进与 context build 生命周期。
边界
| dapei 平台 | Agent |
|---|---|
| stage 状态机、DAG 拓扑排序、context build 流程 | 理解 stage 目标、执行具体任务、填充 context |
| 输出结构化 stage 进度和 context bundle | 理解「这个 stage 需要什么 context」 |
禁止:平台替 Agent 做业务决策或写代码。
强制约束:stage 顺序是 MANDATORY,不是建议。
- analyze-current-state → gap-analysis → solution-design → task-breakdown → implementation → local-validation → architecture-review → acceptance 的依赖是铁律
- "折中方案"、"快速确认"、"部分完成" 都等于违反 DAG
- 任何跳阶段的行为都需要正式的回滚流程,不是"这次例外"
路由能力
| 意图 | Capability |
|---|---|
| 运行单个 stage | workflow.runStage |
| 查看 workflow 状态 | workflow.status |
| 构建 context bundle | context.build |
工作流(方法,非实现)
Stage DAG(阶段有向无环图)
标准 stage 顺序:
analyze-current-state → gap-analysis → solution-design → task-breakdown → implementation → local-validation → architecture-review → acceptance
每个 stage 依赖前一个 stage 完成:
stages:
- name: analyze-current-state
depends_on: []
status: completed
- name: gap-analysis
depends_on: [analyze-current-state]
status: in_progress
- name: solution-design
depends_on: [gap-analysis]
status: pending
- name: task-breakdown
depends_on: [solution-design]
status: pending
- name: implementation
depends_on: [task-breakdown]
status: pending
- name: local-validation
depends_on: [implementation]
status: pending
- name: architecture-review
depends_on: [local-validation]
status: pending
- name: acceptance
depends_on: [architecture-review]
status: pending
runStage(执行阶段)
- 验证依赖 stage 已完成
- 检查 feature 状态允许此 stage 执行
- 按 DAG 顺序执行 stage checklist
- 更新 feature.yaml 中 stage 状态
- 持久化 memory 到
context/memory/ - 更新 progress report
runStage 校验规则:
solution-design/implementation/acceptance必须有用户确认implementation之前必须完成task-breakdownlocal-validation之前必须完成implementationacceptance之前必须完成architecture-review
context.build(构建上下文)
按 stage 加载层叠 context:
L1: Workspace Context — workspace 全局信息
- workspace.yaml 元数据
- repos 列表和状态
L2: Feature Context — feature 级别信息
- feature.yaml
context/memory/历史记录
L3: Stage-specific Context — 按 stage 特化
analyze-current-state: cognitive.discover 输出与 as-is 证据solution-design:docs/03-business-design.md和docs/04-technical-design.mdimplementation:docs/05-task-breakdown.md与 repo worktreelocal-validation:reports/validation-report.md
Cognitive Assets Summary (v2.1) — 每个 stage 输出末尾注入的「资产盘点」段
discover阶段(analyze-current-state/gap-analysis): 显示 profiles / confirmed / candidate entries 计数design阶段(solution-design/task-breakdown/implementation): 显示 behaviors / state machines / business rules 计数ship阶段(local-validation/architecture-review/acceptance): 显示 domains / capability map / docs portal 状态- 空 workspace 时: 给
@dapei cdr bootstrap <repo>的引导
L4: Runtime Context — 运行时聚合
context/runtime-context.md(最终输出)
Memory(记忆持久化)
每个 stage 完成后,记录:
- stage: solution-design
completed_at: 2025-05-20T14:30:00Z
duration: 2h 15m
checkpoints:
- "Reviewed repo structure"
- "Identified 3 behavior entries"
- "Designed state machine for Order"
next_steps:
- "Start implementation stage"
- "Focus on payment flow"
用户入口
@dapei run workflow my-feature --stage implementation
@dapei context build my-feature --stage solution-design
@dapei workflow status my-feature
常见错误
| 错误 | 后果 |
|---|---|
| "用户说跳过就跳过" | 违反 DAG 约束,context 丢失上游信息 |
| "之前做过了,直接标记完成" | 实际没走确认流程,artifact 不完整 |
| "折中方案:只做轻量确认" | 确认点不能协商,轻量确认 ≠ 完成 |
| implementation → analyze-current-state → solution-design 回跳 | DAG 只允许向前推进,不允许回跳 |
红线 — 禁止行为
- 禁止跳阶段或"部分完成"阶段
- 禁止用用户要求作为跳阶段的理由
- 禁止"折中"确认点 — 确认点要么完成要么不完成
- 禁止在未完成前置 stage 的情况下运行后续 stage
Rationalization 堵口
| 借口 | 反驳 |
|---|---|
| "用户说跳过" | 用户无权覆盖 DAG 约束,这由平台强制 |
| "已经做过类似的" | 没走正式流程就不能标记完成 |
| "折中方案够好了" | "够好"等于违反 MANDATORY 约束 |
| "时间紧,先上线再说" | 时间压力不能作为跳流程的理由 |
与其他 skill 的协作
- cognitive:analyze-current-state 阶段使用 cognitive skill 生成 behavior 文档
- repos:workflow 执行依赖 repos 的 worktree 状态
- validation:validate stage 使用 validation skill 执行测试
- feature:workflow 操作 feature.yaml 的 stage 状态