intentdna 1.5.6 → 1.5.7
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/package.json
CHANGED
|
@@ -0,0 +1,101 @@
|
|
|
1
|
+
# Spec: Workflow Pipeline + 状态管理统一
|
|
2
|
+
|
|
3
|
+
## 问题
|
|
4
|
+
|
|
5
|
+
1. **无 workflow 链接**: behavior-lock 完成后不能自动触发 rescue，用户手动跑
|
|
6
|
+
2. **状态管理双路径**: hook CLI 和 MCP server 各自读写同一套文件，重复实现 + 并发风险
|
|
7
|
+
3. **无 pipeline 编排**: 不支持 skill → skill 的串联执行
|
|
8
|
+
|
|
9
|
+
## 一、Workflow Pipeline
|
|
10
|
+
|
|
11
|
+
### 参考 OMC
|
|
12
|
+
|
|
13
|
+
OMC 有两层 pipeline：
|
|
14
|
+
- SKILL.md frontmatter: `next-skill: omc-plan` + `pipeline: [deep-interview, omc-plan, autopilot]`
|
|
15
|
+
- Autopilot 状态机: `STAGE_ORDER: ["ralplan", "execution", "ralph", "qa"]`
|
|
16
|
+
|
|
17
|
+
### DNA 设计
|
|
18
|
+
|
|
19
|
+
在 workflow YAML 层面支持 `next_workflow`:
|
|
20
|
+
|
|
21
|
+
```yaml
|
|
22
|
+
workflows:
|
|
23
|
+
behavior-lock:
|
|
24
|
+
name: Behavior Lock
|
|
25
|
+
next_workflow: rescue # 完成后建议/触发 rescue
|
|
26
|
+
steps: [...]
|
|
27
|
+
|
|
28
|
+
rescue:
|
|
29
|
+
name: Rescue
|
|
30
|
+
max_rounds: 10
|
|
31
|
+
steps: [...]
|
|
32
|
+
```
|
|
33
|
+
|
|
34
|
+
编译到 SKILL.md 的步骤末尾：
|
|
35
|
+
|
|
36
|
+
```markdown
|
|
37
|
+
<Next_Workflow>
|
|
38
|
+
behavior-lock completed. Next recommended workflow:
|
|
39
|
+
Skill("dna-frw-rescue", args="$ARGUMENTS")
|
|
40
|
+
Ask user: "Behavior lock complete. Run rescue to implement? (Y/n)"
|
|
41
|
+
</Next_Workflow>
|
|
42
|
+
```
|
|
43
|
+
|
|
44
|
+
### 编译器改动
|
|
45
|
+
|
|
46
|
+
`skill-adapter.ts` 新增：
|
|
47
|
+
- 读 `WorkflowDef.next_workflow`
|
|
48
|
+
- 生成 `<Next_Workflow>` section
|
|
49
|
+
- 包含 `Skill()` 调用模板
|
|
50
|
+
|
|
51
|
+
### schema 改动
|
|
52
|
+
|
|
53
|
+
```typescript
|
|
54
|
+
interface WorkflowDef {
|
|
55
|
+
// 现有字段...
|
|
56
|
+
next_workflow?: string; // 完成后推荐的下一个 workflow
|
|
57
|
+
auto_chain?: boolean; // true = 自动执行，false = 提示用户确认（默认 false）
|
|
58
|
+
}
|
|
59
|
+
```
|
|
60
|
+
|
|
61
|
+
## 二、状态管理统一
|
|
62
|
+
|
|
63
|
+
### 当前双路径
|
|
64
|
+
|
|
65
|
+
```
|
|
66
|
+
Hook CLI (dna-hook): 读文件 → enforce → 写 trace → 退出
|
|
67
|
+
MCP Server (dna-mcp): 读文件 → 响应工具调用 → 写文件
|
|
68
|
+
两者独立，读写同一套 .dna/state/ 文件
|
|
69
|
+
```
|
|
70
|
+
|
|
71
|
+
### 短期方案（不做 IPC）
|
|
72
|
+
|
|
73
|
+
保持双路径，但统一底层：
|
|
74
|
+
1. **hook CLI 和 MCP 共享状态读写模块** — 已经如此（都用 state.ts）
|
|
75
|
+
2. **trace 写入加 session 隔离** — 已实现（G1）
|
|
76
|
+
3. **workflow state 原子写入** — 已实现（R4）
|
|
77
|
+
4. 当前无需改动，架构够用
|
|
78
|
+
|
|
79
|
+
### 长期方案（等 CC 支持 MCP hooks）
|
|
80
|
+
|
|
81
|
+
```
|
|
82
|
+
Claude Code → hook 事件 → 调用 MCP tool `dna_enforce` → 常驻进程
|
|
83
|
+
- 零冷启动（IR 已缓存）
|
|
84
|
+
- 单点状态管理
|
|
85
|
+
- 无并发写入
|
|
86
|
+
```
|
|
87
|
+
|
|
88
|
+
等 Claude Code hook 系统支持 MCP tool 调用时迁移。
|
|
89
|
+
|
|
90
|
+
## Acceptance Criteria
|
|
91
|
+
|
|
92
|
+
### Pipeline
|
|
93
|
+
- [ ] WorkflowDef 支持 `next_workflow` 字段
|
|
94
|
+
- [ ] Schema 验证 next_workflow 引用存在
|
|
95
|
+
- [ ] skill-adapter.ts 生成 `<Next_Workflow>` section
|
|
96
|
+
- [ ] 默认提示用户确认，`auto_chain: true` 自动执行
|
|
97
|
+
- [ ] flutter-rewrite: behavior-lock → rescue 链接
|
|
98
|
+
|
|
99
|
+
### 状态统一
|
|
100
|
+
- [ ] 记录长期方案到 ROADMAP（MCP-based hooks）
|
|
101
|
+
- [ ] 短期无需改动，确认当前架构够用
|