← Back to Skills Marketplace
denghuayuan

Multica Sdd Workflow

by denghuayuan · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ⚠ suspicious
36
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install multica-sdd-workflow
Description
SDD + Multica 多智能体开发工作流。当用户提到"新建任务"、"发布给 multica"、"创建小队"、 "分配 issue"、"开始新功能"、"组建小队执行"、"按 SDD 流程"、"新增功能"、"功能更新"时触发。 覆盖从 SDD 规范文档编写、multica 小队创建、Issue 发布执行,到完成...
README (SKILL.md)

Multica SDD 工作流

每次新增或更新功能,必须留有完整的 SDD 记录。流程如下:

0. 编写 SDD 规范文档(开工前必须)
       ↓
1. 读取项目配置 + 确认小队
       ↓
2. 创建父 Issue + 子 Issue + Leader Kickoff
       ↓
3. 依赖链规划
       ↓
4. 监控进度
       ↓
5. 完成后更新 SDD 记录(收尾必须)

Step 0:编写 SDD 规范文档(开工前必须)

宪法 Article II § 2.2(API First)+ Article IX § 9.2(Required Documentation)

任何功能,必须先写规范文档,再创建 Issue,再执行。

0.0 模式判断

检测 specs/ 是否已有规范文档:

  • specs/ 目录 → 进入 0-A 新建模式
  • 已有 specs/ 目录 → 进入 0-B 续接模式

0-B 续接模式

目标:不重写已有规范,只补全缺失信息,确保任何 agent 能直接接手。

1. 扫描未完成任务:读取所有 specs/*/tasks.md,统计 [ ] 条目,了解当前进度。

2. 补全格式缺口:对每个未完成的 Task,确认它有 StepsAcceptance Criteria。缺失的就地补全,不新建文件。格式要求见下方 0-A 的 tasks.md 规范。

3. 更新 CLAUDE.md:将「下一步行动」章节精确同步到当前未完成任务,按依赖顺序列出,标注 TDD 阶段(RED/GREEN)和目标文件路径。

完成后直接进入 Step 1,基于现有 specs 继续创建 Issue。tasks.mdCLAUDE.md 是唯一真相来源,无需生成额外文档。


0-A 新建模式:编写全新规范

文档目录结构

specs/
└── \x3CNNN>-\x3Cfeature-name>/     # NNN 三位数字序号,如 002-ipc-handlers
    ├── spec.md               # 需求规范
    ├── plan.md               # 技术方案
    └── tasks.md              # 任务清单

三个文档的格式规范见 sdd-templates.md


Step 1:读取项目配置

在执行任何后续步骤前,必须先从 .multica/squad.md 读取配置。

PROJECT_ID=$(grep '| project-id'       .multica/squad.md | cut -d'`' -f2)
SQUAD_ID=$(grep '| squad-id'           .multica/squad.md | cut -d'`' -f2)
LEADER_ID=$(grep '| leader-id'         .multica/squad.md | cut -d'`' -f2)
EXECUTOR_ID=$(grep '| executor-id'     .multica/squad.md | cut -d'`' -f2)
EXECUTOR_NAME=$(grep '| executor-name' .multica/squad.md | cut -d'`' -f2)
WORK_DIR=$(grep '| work-dir'           .multica/squad.md | cut -d'`' -f2)
TEST_CMD=$(grep '| test-cmd'           .multica/squad.md | cut -d'`' -f2)
TYPECHECK_CMD=$(grep '| typecheck-cmd' .multica/squad.md | cut -d'`' -f2)

# 验证——任何变量为空说明 squad.md 格式有误
echo "PROJECT_ID:    $PROJECT_ID"
echo "SQUAD_ID:      $SQUAD_ID"
echo "LEADER_ID:     $LEADER_ID"
echo "EXECUTOR_ID:   $EXECUTOR_ID"
echo "EXECUTOR_NAME: $EXECUTOR_NAME"
echo "WORK_DIR:      $WORK_DIR"
echo "TEST_CMD:      $TEST_CMD"
echo "TYPECHECK_CMD: $TYPECHECK_CMD"

确认小队在线

multica agent list --output json | grep -E '"name"|"status"'
multica squad list --output json

新项目初始化(首次使用时)

如果项目尚未有小队,执行以下步骤一次性配置:

# 1. 在 multica 创建项目
multica project create --name "\x3C项目名>" --output json
# 记录返回的 project-id → 填入 .multica/squad.md

# 2. 创建小队
multica squad create --name "\x3Csquad-name>" --description "\x3Cdesc>" \
  --leader "$LEADER_ID" --output json
# 记录返回的 squad-id → 填入 .multica/squad.md

# 3. 添加执行 agent 到小队
multica squad member add "$SQUAD_ID" --member-id "$EXECUTOR_ID" --role member

.multica/squad.md 配置表格式(8 个字段,字段名固定):

| 字段 | 值 |
|------|---|
| project-id    | `\x3Cmultica-project-uuid>` |
| squad-id      | `\x3Cmultica-squad-uuid>` |
| leader-id     | `\x3Cleader-agent-uuid>` |
| executor-id   | `\x3Cexecutor-agent-uuid>` |
| executor-name | `\x3CEXECUTOR-AGENT-NAME>` |
| work-dir      | `\x3C项目绝对路径>` |
| test-cmd      | `\x3C测试命令,如 npm test>` |
| typecheck-cmd | `\x3C类型检查命令,如 npm run typecheck>` |

Step 2:创建 Issue

2.1 父 Issue(Phase 级别)

multica issue create \
  --project "$PROJECT_ID" \
  --title "Phase N: \x3C阶段名>(Tasks N.1-N.M)" \
  --description-file /tmp/phase_desc.txt \
  --assignee-id "$SQUAD_ID" \
  --priority high \
  --output json

父 Issue、子 Issue、Leader Kickoff 三个 description 模板见 issue-templates.md


Step 3:依赖链规划

根据 tasks.md 中的 Dependencies 字段,将任务分批派发:

无依赖任务 → 第1批并行
    ↓(第1批全部 in_review 或 done 后触发)
依赖第1批的任务 → 第2批并行
    ↓
汇总任务(集成测试、公共 API 导出等)→ 最后单独一批

示例:Phase N 有 6 个子 Issue:

WS-01(Task N.1 RED,  无依赖)  ─┐
WS-02(Task N.3 RED,  无依赖)   ├─ 第1批并行派发
WS-03(Task N.5 RED,  无依赖)  ─┘
         ↓
WS-04(Task N.2 GREEN, 依赖 WS-01)─┐
WS-05(Task N.4 GREEN, 依赖 WS-02)  ├─ 第2批并行派发
WS-06(Task N.6 GREEN, 依赖 WS-03)─┘
         ↓
WS-07(集成测试, 依赖 WS-04~06)────── 第3批(单独)

划分规则:

  • RED 任务(写测试)先于其对应的 GREEN 任务(写实现)派发
  • 多个互不依赖的任务同批并行
  • 有多个前置依赖的任务,等所有前置均达到 in_reviewdone 后再派

Step 4:监控进度

# 先加载配置(新 shell 时)
PROJECT_ID=$(grep '| project-id' .multica/squad.md | cut -d'`' -f2)
LEADER_ID=$(grep '| leader-id'   .multica/squad.md | cut -d'`' -f2)

# 查看所有 Issue 状态
multica issue list --project "$PROJECT_ID" --output json \
  | grep -E '"identifier"|"title"|"status"'

# 查看单个 Issue
multica issue get \x3CWS-XX>

# Leader 无响应时重新触发
multica issue assign \x3Ckickoff-id> --to "$LEADER_ID"

状态流转:todoin_progressin_reviewdone


Step 5:完成后更新 SDD 记录(收尾必须)

宪法 Article IX § 9.1(Living Documentation)+ Article XI § 11.1(Git Workflow)

Phase 完成后,必须同步更新以下记录,再提交。

5.1 更新 tasks.md

将完成的任务 checkbox 从 [ ] 改为 [x],并统计完成情况:

# 确认还有哪些未完成
grep -n "\[ \]" specs/\x3CNNN>-\x3Cfeature>/tasks.md

# 统计完成进度
DONE_COUNT=$(grep -c "\[x\]" specs/\x3CNNN>-\x3Cfeature>/tasks.md)
TOTAL_COUNT=$(grep -c "^### Task" specs/\x3CNNN>-\x3Cfeature>/tasks.md)
echo "完成进度: $DONE_COUNT / $TOTAL_COUNT"

5.2 更新 CLAUDE.md 进度

## 开发进度

**总进度**: \x3C已完成>/\x3C总任务> 任务完成

### Phase N: \x3C阶段名> ✅ 完成(完成数/本阶段总数 任务)

| 任务 | 状态 | 描述 |
|------|------|------|
| N.1  | ✅   | ...  |
| N.2  | ✅   | ...  |

**已完成成果**:
- ✅ \x3C文件或功能点>

CLAUDE.md 需更新的字段:

  • **项目状态** 百分比
  • **总进度** 任务数(从 tasks.md 动态统计,不要写死)
  • 对应 Phase 的状态从 🔄 进行中 改为 ✅ 完成
  • **最后更新** 日期

5.3 提交记录更新

DONE_COUNT=$(grep -c "\[x\]" specs/\x3CNNN>-\x3Cfeature>/tasks.md)
TOTAL_COUNT=$(grep -c "^### Task" specs/\x3CNNN>-\x3Cfeature>/tasks.md)

git add specs/\x3CNNN>-\x3Cfeature>/tasks.md CLAUDE.md
git commit -m "docs(sdd): mark Phase N complete, update progress to ${DONE_COUNT}/${TOTAL_COUNT}"

5.4 更新 .multica/squad.md Issue 历史

.multica/squad.md 的 Issue 历史清单中追加本次新建的 Issue:

| WS-XX | Phase N: \x3C阶段名>(父) | — | ✅ done |
| WS-YY | Task N.X+N.Y: \x3C功能>   | WS-XX | ✅ done |

关键约束(来自 AGENTS.md + 宪法)

约束 来源
TDD:RED 先提交,GREEN 后提交 宪法 Article III
禁止 git pushgit add . AGENTS.md
提交格式:\x3Ctype>(\x3Cscope>): \x3Cdesc> (Task N.X RED/GREEN) AGENTS.md
生产依赖 ≤ 10 个 宪法 Article IV § 4.2
TypeScript strict 零错误 宪法 Article V § 5.2
启动 \x3C 2s,操作 \x3C 100ms 宪法 Article VI § 6.1
零网络调用 宪法 Article VII § 7.1
单元测试覆盖率 ≥ 80% 宪法 Article III § 3.2

注意:以上约束来自 memo-desktop 项目的宪法。其他项目使用此 skill 时,应根据自身 AGENTS.md 和宪法文档替换此约束表。

Usage Guidance
Install only if you intentionally want generic feature-planning requests to route into a Multica SDD workflow. Before using it, narrow the trigger wording or require explicit confirmation before creating or assigning Multica issues, editing specs/CLAUDE.md/.multica/squad.md, or making git commits.
Capability Assessment
Purpose & Capability
The requested Bash, file-read, and Multica issue/squad commands fit the stated SDD + Multica workflow, including creating specs, creating issues, assigning agents, monitoring status, and updating records.
Instruction Scope
The trigger text includes generic feature-work phrases such as new task, new feature, and feature update, while the workflow then says to proceed through issue creation, squad assignment, file updates, and commits. That creates real risk of invocation outside a clearly intended Multica workflow.
Install Mechanism
The artifact contains only Markdown skill and template files; no executable installer, package dependency, or hidden install-time script was present.
Credentials
The skill mainly scopes work to project files such as specs, CLAUDE.md, and .multica/squad.md, but it also uses Multica CLI commands that can create projects, squads, issues, assignments, and status updates in an external system.
Persistence & Privilege
It instructs the agent to persist workflow state by updating project documentation, appending issue history, and making git commits, and it coordinates other agents through Multica without requiring explicit user confirmation before those state changes.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install multica-sdd-workflow
  3. After installation, invoke the skill by name or use /multica-sdd-workflow
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
Version 2.0.0 – Major update: introduces a complete, documented workflow for SDD + Multica multi-agent collaborations. - Adds comprehensive step-by-step workflow for feature development, from SDD documentation to squad/issue management and progress tracking. - Strictly enforces SDD-first, TDD (RED/GREEN) process, and documentation updates at every phase. - Details squad configuration, initialization, and real-time progress monitoring. - Clarifies issue creation, dependency management, and reporting requirements. - Codifies all operational and constitutional project constraints. - Includes actionable bash snippets and template references for each workflow stage.
Metadata
Slug multica-sdd-workflow
Version 1.0.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 1
Frequently Asked Questions

What is Multica Sdd Workflow?

SDD + Multica 多智能体开发工作流。当用户提到"新建任务"、"发布给 multica"、"创建小队"、 "分配 issue"、"开始新功能"、"组建小队执行"、"按 SDD 流程"、"新增功能"、"功能更新"时触发。 覆盖从 SDD 规范文档编写、multica 小队创建、Issue 发布执行,到完成... It is an AI Agent Skill for Claude Code / OpenClaw, with 36 downloads so far.

How do I install Multica Sdd Workflow?

Run "/install multica-sdd-workflow" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Multica Sdd Workflow free?

Yes, Multica Sdd Workflow is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Multica Sdd Workflow support?

Multica Sdd Workflow is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Multica Sdd Workflow?

It is built and maintained by denghuayuan (@denghuayuan); the current version is v1.0.0.

💬 Comments