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:

• - INLINECODE0
• INLINECODE1
• INLINECODE2
• INLINECODE3

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. 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
• INLINECODE5 value and resource bot
• INLINECODE6 problem definition and logic bot
• INLINECODE7 long-term direction and principles bot
• INLINECODE8 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

• - INLINECODE9
• INLINECODE10
• INLINECODE11
• INLINECODE12
• INLINECODE13
• INLINECODE14
• INLINECODE15

L5 docs

• - INLINECODE16
• INLINECODE17
• INLINECODE18
• INLINECODE19
• INLINECODE20
• INLINECODE21

Task system

• - INLINECODE22
• INLINECODE23
• INLINECODE24

Per-bot files

For 001 to 005, create:

• - INLINECODE27
• INLINECODE28
• INLINECODE29
• INLINECODE30

---

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
• INLINECODE35 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. 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
• INLINECODE37 for final quality checks
• INLINECODE38 for task card / blackboard / review minimal examples
• INLINECODE39 for quick trigger and delivery checks and realistic request coverage
• INLINECODE40 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.