openclaw-multi-agent-hq-builder（OpenClaw multi Agent HQ 构建器）

OpenClaw Multi-Agent HQ Builder

Skill purpose

Use this skill to help a user build a mother-bot + sub-bots OpenClaw organization that is:

• structured
• installable
• teachable
• runnable
• upgradeable

When speed matters, optimize for a newcomer being able to finish the installation with a clear checklist, exact file list, and acceptance criteria before the stated deadline.

This skill is not for random prompt roleplay. It is for turning a multi-agent idea into a working OpenClaw operating system skeleton.

What this skill builds

A standard delivery should include these layers:

1. Organization layer

Create or refine:

• org chart
• mother-bot / sub-bot division
• role boundaries
• reporting lines

2. Agent layer

For each core bot, create the four-file profile:

• SOUL.md
• HEARTBEAT.md
• ROLE.md
• AGENT_PROFILE.md

3. Protocol layer

Create or refine:

• work handbook
• task flow diagram
• input/output templates
• dispatcher rules

4. L5 operations layer

Create or refine:

• task state machine
• task card template
• blackboard protocol
• review mechanism
• system upgrade log

5. Onboarding layer

Create or refine:

• installation guide
• newcomer checklist
• first task demo card

Default build order

Do not start by writing lots of personalities. Build in this order:

1. Confirm the mother-bot and core sub-bots
2. Confirm role boundaries and responsibilities
3. Build the HQ docs
4. Install each bot's four-file profile
5. Build dispatcher and workflow docs
6. Build task state machine and blackboard docs
7. Build onboarding docs for newcomers
8. Create a first live task card example

Reason: new users fail when they start with too many agents and too little structure.

Recommended core architecture

Unless the user clearly wants a different structure, prefer this minimum HQ:

• 001 mother-bot / CEO / dispatcher / final decider
• 02 value and resource bot
• 03 problem definition and logic bot
• 04 long-term direction and principles bot
• 05 execution and delivery bot

Keep the HQ small and hard. If the user mentions 06-10 or more specialized agents, default to an outer specialist pool, not permanent HQ members.

Standard delivery files

When building from scratch, aim to produce these files or their equivalents:

HQ docs

• 组织架构总表.md
• 母bot与子bot工作手册.md
• 001-05任务流转图.md
• 001-05标准输入输出模板包.md
• 001-dispatcher.md
• 001-dispatcher-runbook.md
• 多智能体技术路线图.md

L5 docs

• 任务状态机.md
• 任务卡模板.md
• 共享黑板协议.md
• 复盘机制.md
• 复盘模板.md
• 系统升级日志.md

Task system

• tasks/
• tasks/TASK-001-*.md
• tasks/TASK-002-*.md

Per-bot files

For 001 to 005, create:

• teammates/<id>/SOUL.md
• teammates/<id>/HEARTBEAT.md
• teammates/<id>/ROLE.md
• teammates/<id>/AGENT_PROFILE.md

Installation workflow for a newcomer

When the user wants a teachable installation package, produce an onboarding guide that walks them through:

Step 1: Confirm structure

Have them decide:

• how many core bots
• what each bot is responsible for
• what stays in HQ vs outer specialist pool

Step 2: Build HQ documents

Create the shared HQ governance files first.

Step 3: Install the bot profiles

Create the four-file profile for each bot.

Step 4: Install the dispatcher

Set up 001-dispatcher.md and 001-dispatcher-runbook.md.

Step 5: Install L5 operations

Set up state machine, task cards, blackboard, review system, and upgrade log.

Step 6: Run one real task

Create a first real TASK-001 or TASK-002 and run the full chain.

Newcomer success checklist

A newcomer should be considered successful only if all of the following are true:

• HQ structure is clear
• each core bot has all 4 files
• dispatcher exists
• task state machine exists
• task card template exists
• blackboard protocol exists
• review mechanism exists
• upgrade log exists
• tasks/ directory exists
• at least one real task card exists
• the user can explain the routing chain from 001 to sub-bots and back

Output style

When using this skill, prefer outputs that are:

• concise
• structured
• installable
• easy to follow for a newcomer
• explicit about file paths and check steps

Do not give abstract theory without telling the user what files to create and in what order.

Suggested report structure

For setup requests, use this structure:

1. Conclusion
2. What will be created
3. Exact file list
4. Recommended order
5. Key risks
6. Final acceptance checklist

Anti-patterns

Avoid these mistakes:

• too many bots too early
• vague role descriptions with no file outputs
• no dispatcher
• no state machine
• no task cards
• no review loop
• no newcomer checklist
• relying only on chat history instead of files

Bundled resources

Read these bundled resources when needed:

• references/install-guide.md for newcomer installation teaching
• references/acceptance-checklist.md for final quality checks
• references/minimal-examples.md for task card / blackboard / review minimal examples
• references/test-prompts.md for quick trigger and delivery checks and realistic request coverage
• references/publish-checklist.md for pre-release checks before declaring the skill ready to ship

Core sentence

The goal is not to create more agents. The goal is to create a small, hard, teachable multi-agent HQ that can actually run.