A comprehensive framework for architecting and managing specialized AI agents with independent sessions, personalities, and sandboxed environments.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install openclaw-agent-builder
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install openclaw-agent-builder using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is OpenClaw Agent Builder?
OpenClaw Agent Builder provides the essential infrastructure to create and configure AI agents within the OpenClaw ecosystem. By leveraging a workspace-based architecture, each agent maintains its own SOUL (persona), AGENTS (workflow instructions), and TOOLS (permission sets). This modular approach ensures that Openclaw Skills can be deployed across various channels like Feishu, WhatsApp, or Discord with precise routing and isolation rules, allowing for a highly scalable and secure multi-agent environment.
OpenClaw Agent Builder Use Cases
- Deploying specialized agents for domain-specific tasks such as finance, development, or supervisor roles.
- Isolating private conversation histories in shared communication channels using dmScope.
- Synchronizing user identity across different messaging platforms like Telegram and Slack via identityLinks.
- Running untrusted code safely using Docker-based sandbox environments.
- Automating routine checks and notifications through the built-in heartbeat mechanism for Openclaw Skills.
How OpenClaw Agent Builder Works
- The system initializes a dedicated workspace directory containing core behavioral definitions in markdown format.
- The developer registers the agent via the CLI, mapping it to a specific workspace and model configuration.
- Bindings are established to route incoming messages from specific channels or users to the correct agent ID.
- When a message is received, the agent loads the session context, applies persona rules from SOUL.md, and consults AGENTS.md for workflow logic.
- The agent executes permitted tools or Openclaw Skills, interacting with the file system or external APIs.
- Session history is persisted in JSONL format, ensuring continuity and long-term memory retrieval.
OpenClaw Agent Builder Setup
To begin building with Openclaw Skills, initialize your workspace and register your agent:
# Create a workspace directory
mkdir -p ~/openclaw-workspaces/my-agent
cd ~/openclaw-workspaces/my-agent
# Initialize the workspace structure
openclaw setup --workspace .
# Create required configuration files
touch AGENTS.md SOUL.md TOOLS.md USER.md
# Register the agent with the OpenClaw system
openclaw agents create my-agent --workspace ~/openclaw-workspaces/my-agent
# Set the model provider
openclaw agents config my-agent --model anthropic/claude-3-5-sonnet
OpenClaw Agent Builder Data Schema & Taxonomy
The agent builder organizes data using a structured filesystem hierarchy to ensure clarity and portability of your Openclaw Skills:
| Directory/File | Description | Format |
|---|---|---|
openclaw.json |
Global configuration and agent registry | JSON |
agents/<id>/sessions/ |
Per-agent session history storage | JSONL |
workspace/SOUL.md |
Personality, tone, and identity definitions | Markdown |
workspace/AGENTS.md |
Core workflow instructions and safety rules | Markdown |
workspace/TOOLS.md |
Tool integration details and local settings | Markdown |
workspace/memory/ |
Daily context and short-term logs | Markdown |
OpenClaw Agent Builder Advanced Features
- Multi-agent routing via Bindings for precise message filtering based on channel or peer ID.
- Session maintenance policies for automatic pruning and rotation of large JSONL history files.
- Granular tool permissions using allow and deny lists to restrict agent actions in sensitive environments.
- Support for custom lifecycle hooks (bootstrap, message_received, etc.) to trigger external scripts.
- Cross-channel identity linking to maintain a single conversation thread for one user across multiple platforms.
- Heartbeat mechanisms to enable proactive agent actions and routine check-ins for Openclaw Skills.
SKILL.md
Loading
Related Openclaw Skills
Eywa
A sophisticated multi-agent coordination layer providing shared spatial memory and swarm navigation for distributed AI agents.
MinerU PDF Extractor
A high-performance PDF-to-Markdown extraction tool supporting OCR, formulas, and complex table layouts via the MinerU API.
Excalidraw Diagram Generator
Transform text descriptions into beautiful, hand-drawn diagrams and architecture charts rendered as high-quality PNG images.
TS API Miner
An intelligent assistant for analyzing TypeScript files to extract API interfaces, request structures, and JSDoc comments into structured Markdown documentation.
Code Quality Analyzer
A professional-grade static code analysis tool designed to detect bugs, assess complexity, and enforce best practices across multiple programming languages.
Command Creator
A specialized tool for generating and organizing custom Claude Code slash commands to automate repetitive development tasks.
Featured*
