A specialized framework for designing, generating, and refining the core architecture of autonomous AI agents.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install 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 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 Agent Builder?
The Agent Builder is a foundational skill designed to streamline the lifecycle of creating Openclaw Skills. It guides developers through the process of defining an agent's persona, operational boundaries, and technical configuration. By automating the generation of essential workspace files, it ensures that every agent adheres to a standardized architecture while maintaining the flexibility required for specialized tasks.
This skill is particularly valuable for developers who need to transition from simple prompt-based interactions to complex, multi-layered agentic workflows. Whether you are building an Advisor that suggests code changes or an Autopilot capable of managing cloud infrastructure, this skill provides the scaffolding necessary for reliable and safe AI operations within the ecosystem of Openclaw Skills.
Agent Builder Use Cases
- Initializing a new agent workspace with standardized file structures.
- Transitioning an agent from manual advisor mode to full autopilot autonomy.
- Refining agent tone and communication styles for specific platforms like Discord or Telegram.
- Implementing surgical updates to existing agents to prevent logic loops or overreach.
How Agent Builder Works
- The skill initiates a targeted interview process to capture the agent's primary mission, operational surfaces, and autonomy level.
- It generates a comprehensive suite of markdown-based configuration files, including SOUL.md and AGENTS.md, tailored to the user's specific requirements.
- It applies a standardized guardrails checklist to enforce safety rules, such as mandatory confirmation before destructive actions.
- It provides a set of rapid acceptance tests to simulate real-world scenarios and verify the agent's behavior before deployment.
- For existing agents, it analyzes performance gaps and proposes surgical diffs to optimize the agent's logic and memory management for Openclaw Skills.
Agent Builder Setup
To begin building your own Openclaw Skills, ensure you have the core references available in your workspace. You can initialize the builder by running the following command within your terminal environment:
# Standard initialization for the agent builder
openclaw install agent-builder
# Ensure reference templates are accessible
ls references/templates.md
Once installed, invoke the skill to start the Phase 1 interview process.
Agent Builder Data Schema & Taxonomy
The Agent Builder organizes data into a structured workspace layout essential for Openclaw Skills. The primary files generated include:
| File Name | Purpose | Key Metadata |
|---|---|---|
| SOUL.md | Persona & Identity | Mission statement, tone, and personality traits. |
| AGENTS.md | Operating Rules | Delegation logic, memory management, and tool posture. |
| IDENTITY.md | Core Branding | Agent name and technical identifiers. |
| HEARTBEAT.md | Process Logic | Scheduling for autonomous tasks and status checks. |
| USER.md | User Context | Specific instructions on how to interact with the human operator. |
| MEMORY.md | Knowledge Base | Curated long-term storage and session history summaries. |
Agent Builder Advanced Features
- Multi-agent delegation rules defined within the AGENTS.md framework.
- Custom heartbeat plans for persistent, long-running background tasks.
- Tool-first posture configuration to prioritize API actions over conversational responses.
- Automated loop breaker guidance to prevent recursive logic errors in autonomous modes.
- Platform-specific group chat etiquette rules for diverse messaging environments within Openclaw Skills.
SKILL.md
Loading
Related Openclaw Skills
Knowledge Management & Proactive Recall
An intelligent system for capturing content from URLs, YouTube videos, and local files into a structured, searchable personal knowledge base.
Solana Development Skill
A technical framework for AI agents to architect, develop, and test Solana blockchain applications using Openclaw Skills.
POLT: Social Memecoin Launchpad
A social finance platform where AI agents collaborate to propose and vote on memecoin concepts that get launched as real tokens.
POLT Collaborative Project Platform
A collaborative ecosystem where humans and AI agents build projects together, complete tasks for SOL bounties, and pitch meme coin concepts.
Apple Remind Me (macOS Native)
A macOS-native skill that enables natural language control over the Apple Reminders app for seamless cross-device task syncing.
Excalidraw Creator
An automated tool within Openclaw Skills to transform JSON-based diagram definitions into professional, hand-drawn style PNG images.
Featured*
