OpenClaw deploys flybooks, multiple separate Agent configuration guides

With the windOpenClawI didn't know what to do with him。

Seeing the operating face: how different is this from a normal big model, and you can't work automatically. Your lobster is a regular chat box

I went from pure white, which OpenClaw doesn't understand, to successfully commanding the Lobsters。

Today, the whole plan to make a successful boss in the Flying Book。

It's all off-the-shelf. You just have to copy the paste

Just do it. You can be the boss of flying books

I. The way to break the rules is in it

In OpenClaw's world, there's a lot of Agents in it, so they're specialized。

This article addresses this core issue。

I've managed to be the boss and hired four employees:

• boss - commander-in-chief, responsible for integrated planning and mission distribution
• researcher - intelligence specialist, specializing in information mining
• word - content creator, good at words
• coder - technical work, focus on code development

Each member has a separate “desk” (workspace), “staff card” (certification) and “work log” (remember) and does not interfere with each other。

Preparation: building your first Agent

2.1 Environmental readiness

Take the Mac system, for example, to open the terminal in sequence:

# installation Homebrew (if installed)
/bin/bash -c "$ (curl-fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

# installation Node.js
I'm going to tell you something.

# installation OpenClaw
curl-fsSL https://openclaw.ai/install.sh bash

# authentication installation
Openclaw-version

Seeing the version number means it's working. If there's no response, redo step three。

2.2 Initialization Configuration

Execute initialization orders:

it's an openclaw on board-install-daemon

Operate with wizard tips:

1. select yes to start configuration
2. Select QuickStart Fast Mode
3. Select your preferred AI model (recommended OpenAI/Claude/Gemini)
4. Skip other advanced options temporarily (follow-up adjustable)

Once you've finished, you have the first available

III. Teamwork: Create your Agent employee

3.1 What is the essence of Agent

Think of Agent as an independent employee. For each employee:

• Workspace - Store job code (AGENTS.md), SOUL.md, owner data (USER.md) and work notes (memory/)
• Employee Cards - Save Identification and Login Certificate
• Dialogues - Recording the history of communication with different objects, independent

3.2. The company is so structured

shell

~/.openclaw/
Ideas - Openclaw.json # Company Regulation (Global Profile)
Ideas - Workspace/ # desk for Jiji
│ - AGENTS.md # workflow manual
│ -SOUL.md # Characteristics Set
Ideas - USER.md # Boss Information Files
│ - memory/ # working memory storage
- Workspace-researcher/ # Inspector's desk
AGENTS.md
SOUL.md
–Momoy/
- Workspace-writer/ # desk
AGENTS.md
SOUL.md
–Momoy/
W-workspace-coder/ #
AGENTS.md
SOUL.md
–Momoy/
Agents/
Ideas - boss/agent/ # ID cabinet for Jiji
Ideas-auth-profiles.json
│ └ - sessions/
Identifier-researcher/agent/ # identification cabinet
Identifier - Writer/agent/ #
│-coder/agent/ #
Skills/ # shared skills

3.3 Two recruitment modalities

Mode I: Command-line Wizard (simple operation)

# Recruitment of Researchers
Now, openclaw forces aded searcher

# Recruitment of Writers
Now, openclaw begins addwriter

# Recruitment of Developers
Now, openclaw forces add code

# View Team Members
Now, openclaw agencies list-bindings

The wizard automatically completes all configurations: construction of spaces, opening of cards, setting of permissions。

Mode 2: Manually edit configuration files

directly modify ~/.openclaw/openclaw.json:

3.4 Important Reminders

⚠️ Identity cannot be shared: each Agent must have an independent auth-profiles.json.

⚠️ Workspace is not reusable: different Agent uses the same angentdir triggers a conflict of authentication and must be avoided。

skill skill skills management: skills/ is an exclusive personal skill in the work area, ~/.openclaw/skills/ is a fully shared skill base。

iv. Create Flying Books

4.1 Create Flying Book Bot Steps

1. open platform for login (open.feishu.cn)
2. Create business self-build applications
3. Create an independent application for each Agent: Bot (Chief) Detective Bot Bot (Investigator) The Bot (author) Possibilities Bot
4. Fetching key information for each Bot: App ID App Security Services Token Encrypt Key
5. Enable robotic capabilities, configure permissions: access basic user information, access and update group messages, view group messages, update group messages, get and send group messages, receive group conversations @ I'm just talking, receiving @robot news, reading the user's message to the robot, sending it as an application, getting and uploading pictures or file resources
6. Modify subscription to: "Receive an event with a long link" and add "Receive a message" event。
7. In Version Management and Release, create a version and publish it。

4.2 Configure flight book channels

Configure flight document authentication information in OpenClaw. Execute for each Agent:

# Flying Book with Jiji
I'm sorry, I'm sorry.

# Configuring Flying Book Bot
I don't know what you're talking about.

# Configure Mann Flying Book Bot
I don't know what you're talking about.

# configuration of a flying book of hackers
I don't know what you're talking about

Enter the respective App ID, App Secret information by hint。

4.3 Creation of Flying Book Group

Main workgroup (to communicate directly with Jiji):

• Create a group chat and only pull into the brain, Bot
• This is the main interface for your mission

Internal collaboration group (synergy space between Agent):

• _Other Organiser
• Into Ziqi, detective, Man, and the Four Bot
• You can join the group. You can watch and interrogate

METHOD TO GET GROUP ID:

1. USE "ACCESS TO GROUP INFORMATION" API ON THE OPEN FLYING BOOK PLATFORM
2. View with OpenClaw session_list command

V. Division of labour: assigning Agent to different positions

Team members are in place, and it is now up to Bindings (routing mechanism) to assign them。

5.1 Bindings’ working principles

Bindings is essentially a message distribution rule: deciding which Agent should handle information from different sources。

OpenClaw matches the following priority:

1. 🎯 PRECISION MATCH - SPECIFY SPECIFIC USER ID OR GROUP ID
2. 👥 Group Match — Matches selected group conversations
3. 📱 Account Match - Match specific Bot Account
4. 📡 Channel Matches – Matches all messages from the entire channel
5. 🧩 Default End - Give to Default

5.2 Our actual configuration

We distinguish between groups and accounts:

_Other Organiser
"bindings":
/ The news of the main team
_Other Organiser
"Agent Id": "Boss,"
"metch":
"channel": "feishu",
"peer":
♪ I'm sorry ♪
♪ I don't know ♪
/ / In-house collaboration, different Bot account numbers
_Other Organiser
"Agent Id": "Boss,"
"metch":
"channel": "feishu",
"account Id": "AppID of Aki."
"peer":
♪ I'm sorry ♪
♪ I don't know ♪
_Other Organiser
"Agent Id": "researcher,"
"metch":
"channel": "feishu",
"Account Id": "AppID of Explorer,"
"peer":
♪ I'm sorry ♪
♪ I don't know ♪
_Other Organiser
"Agent Id": "writer,"
"metch":
"channel": "feishu",
"account Id": "AppID in Man,"
"peer":
♪ I'm sorry ♪
♪ I don't know ♪
_Other Organiser
"Agent Id": "coder,"
"metch":
"channel": "feishu",
"account Id": "AppID of the Extortion,"
"peer":
♪ I'm sorry ♪
♪ I'm sorry ♪
_
♪ I'm sorry ♪

VI. Opening the inside: letting Agent talk to each other

The team is ready, the job is divided, but at this point they can't communicate with each other。

6.1 Activate communication between Agent

This function is disabled by default and needs to be manually activated:

_Other Organiser
"tools":
"Agent ToAgent":
"enabled": true,
"allow": ["Boss", "researcher", "writer", "coder"]
♪ I'm sorry ♪
♪ I'm sorry ♪
♪ I'm sorry ♪

6.2 Setting Session Visibility

By default, each Agent can only see its own conversation records。

To achieve teamwork, the visibility needs to be adjusted:

_other organiser
"tools":
"sessions":
"visibility": "all"
♪ i'm sorry ♪
♪ i'm sorry ♪
♪ i'm sorry ♪

Visibility options description:

• "self" — only your own conversation
• "tree" - look at yourself and the child's mission (default)
• "Agent" - Look at all sessions of the same Agent
• "all" -- look at all Agent's sessions

6.3 Method of sending internal messages

Could not close temporary folder: %s

sessions_send()
"sessionKey": "Agent:researcher:feishu:group: collaborative group ID,"
"Message": "Survey, please look at the latest developments in AI Agent."
"timeoutSeconds": 60
I'm sorry

6.4 Two types of assignment

organisation

• Application scene: Agent is online and has active sessions
• Preconditions: need to configure anentToAgent.enabled and sessions.visibility

organisation

• Applicable scenario: one-time, isolated tasks
• Advantages: Target not required Agent already has active sessions
• supports mode=run (one-time) and mode=session (permanent)

Simple analogy:

• sessions_send = texting to a colleague so he can handle it now
• sessions_spahn = temporary hire of an outsourced person to finish off

VII. CHECK: Establishment of a team information centre

Internal communications have been made available, but there is also a problem: Quiz has been assigned to an expedition without the knowledge of the Man and the Racists; and Quiz has completed research reports that others have to wait for Quiz to see。

The solution is to build a shared knowledge base that all members can read and write。

7.1 Create Shared Directory Structure

mkdir-p ~/.openclaw/shared/notes

The suggested catalogue is as follows:

~/.openclaw/shared/
Ideas - board.md # team bulletin board: Important announcements and developments
The --tasks.md # taskboard: current and completed tasks
└ - notes/ # document library
Ideas - Research/# Research
Ideas - draft articles/ #
code/ # code document

7.2 Configure Shared Directory Index

Lets each Agent's memory search cover the shared directory:

_Other Organiser
"agents":
"defaults":
"MemorySearch":
"extraPaths":
♪ I'm sorry ♪
♪ I'm sorry ♪
♪ I'm sorry ♪
♪ I'm sorry ♪

Actual effects:

• explored to save the study to share/notes/research/ai-trends.md
• you can search for this report with a memory_search
• you can also read files directly with absolute path read command

7.3 Regulation of use of shared documents

markdown

# Team Bulletin Board

## Focus this week
- [2026-03-07] Explorer: AI Agent technical research completed
- [2026-03-07] Languages: Technical Blog Writing
- [2026-03-08] Polars: data analysis script development 📅 to start

## Important Notices
- Agent inter-communication system is online and works with each other through sessions_send
- A shared knowledge base has been established and all documents should be centrally stored in share/notes/

markdown

# Task Viewer

##🔄 ongoing
- [2026-03-07] Published in OpenClaw Operational Guide — Published by: Jiji!
- Expected completion: 2026-03-08
- Dependence: investigative research

##✅ Completed
- [2026-03-07] Explorer: AI Agent Technical Research - Sender: Boss.
- Completion: 2026-03-07 18:30
- Output: Shared/notes/research/ai-transends.md

## 📅 to start
- [2026-03-08] Ultrax: Data visualized scripts - Sender: Jiji

7.4 Define norms for teamwork

Add a code of collaboration to the AGENTS.md of each Agent:

The working norms of Jiji (General Command):

• After taking over the boss's mission, analyze and decompose the mission
• to be assigned to the corresponding commissioner through the sessions_send
• update as soon as tasks are assigned
• Up-to-date the status of the mission and report to the boss upon receipt of the completed feedback

Researcher (researcher)

• We'll start the investigation immediately
• the results of the study are shared to share/notes/research/
• update after completion
• Reply to the completion of the mission to Ji-chul

Texts (authors)

• after receiving the writing assignment, consult the relevant information in shared/notes/
• save draft to shared/notes/drafts/
• _other organiser
• Supported by the coordination of investigations with the Zhuji, if additional information is required

The work of the work of the extremists (developers):

• After receiving a development assignment, consult relevant technical documentation
• code and technical files saved to share/notes/code/
• update after completion
• If you need technical research, you'll be assisted by a little twilight

VIII. Factual presentation: full task flow process

Seeing a real scene, the boss asked for a technical article on AI Agent, accompanied by a demonstration script。

8.1 Mandate flow process

Boss.
"I need a technical article on AI Agent Multi-Intelligence Collaboration.
If you need a real case, you'd better have a demonstration script."

"Step two" for Ji-ji's process:
1. Analytical tasks, to be decomposed into three sub-tasks:
- Technical research (detection)
- Writing articles (written)
- Script development (extendors)

2. Updates shared/tasks.md recording tasks

3. Missions to missions:
Sessions_send()
"sessionKey": "Agent:researcher:feishu:group: collaborative group ID,"
"Message": "Survey, study the current state of technology for AI Agent multi-intelligence,
Mainstream programmes and practical cases, reports are kept to share/notes/research/,"
"timeoutSeconds": 120
I'm sorry.

4. Reply to boss: "Roy! A technical study has been arranged to complete the post-language articles.
The hacker will develop a demonstration script. Delivery is expected tomorrow. " ... "

The third step of the expedition:
1. Receiving missions, starting web search and documentation
2. Agent Framework for Mainstreaming Research: OpenClaw, AutoGPT, LangChain, etc.
3. Collating practical application cases
4. Preparation of studies: shared/notes/research/multi-agent-research.md
5. Update of sared/tasks.md tag assignments completed
6. Response to Jiji: "Study completed, report saved to share/notes/research/multi-agent-research.md"

[Step IV] Coordination of the text:
1. Notification of completion of reconnaissance
2. Text-writing missions:
Sessions_send()
"sessionKey": "Agent:writer:feishu:group: Team ID,"
"Message": "Man, based on shared/notes/research/multi-agent-research.md
I wrote a technical article demanding that it be easy to understand, with practical examples.
The draft is saved to share/notes/drafts/."
"timeoutSeconds": 180
I'm sorry.
3. Update

The creation of the text:
1. Access to research reports
2. Structure and content framework of the articles
3. Writing articles, incorporating case studies and analyses
4. Preservation of drafts: Shared/notes/drafts/multi-agent-article.md
5. Update
6. Reply to Jiji: "Preliminary article completed and saved to share/notes/drafts/"

[Six Step] Schedulings:
1. Notification of completion of receipt of correspondence
2. Sending development missions to extremist groups:
Sessions_send()
"sessionKey": "Agent:coder:feishu:group: team ID,"
"message": "Travellers, according to shared/notes/drafts/multi-agent-article.md
AND THE TECHNOLOGY PROBLEMS IN IT,
The code is saved to share/notes/code/,"
"timeoutSeconds": 240
I'm sorry.
3. Update

The development process of the extremists:
1. Read draft articles and understand technical programmes
2. Design script architecture
3. Preparation of presentation codes
4. Save code and description documents: shared/notes/code/multi-agent-demo/
5. Update
6. Replying to Jiji: "Showcasing script development completed, code in shared/notes/code/multi-agent-demo/"

Zhiji reported to his boss:
"The mission is complete!
• Technical research report: Shared/notes/research/multi-agent-research.md
• Draft article: Shared/notes/drafts/multi-agent-article.md
💻 Presentation script: Shared/notes/code/multi-agent-demo/
Please consult and advise if necessary. " ... "

8.2 Core Collaboration Mechanisms

Throughout the process, the boss said one word and the rest was done by the team:

1. Dismantling of tasks - Ji-chul split complex tasks into enforceable sub-missions
2. Parallel collaboration — Commissioners working independently in their own areas
3. information sharing - delivery of results through shared/ directory
4. progress tracking — tasks.md reflects mission status in real time
5. Auto-coordination — Jiji is responsible for task flow and progress monitoring

IX. At the end

The build-up of the Agent team is just the beginning, and the real value lies in its continued optimization and use。

There are also a few recommendations:

1. Start with simple tasks — let the team handle simple tasks and gradually increase complexity
2. Continuous Optimization of Human Sets - Adjustment of SOUL.md and AGENTS.md to Actual Performance
3. create a working code - define the working code in shared/board.md
4. Regular reviews - allowing Ziqi to regularly take stock of issues and points of improvement in team collaboration
5. good use of a shared knowledge base — depositing valuable information into a shared/directive to form a team knowledge asset