A professional-grade framework for designing and generating complete, production-ready Openclaw Skills agent workspaces from scratch.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install marjorie-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 marjorie-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 specialized tool designed for developers and power users looking to construct robust AI personas within the Openclaw Skills framework. It moves beyond simple prompt engineering by automating the creation of a full workspace architecture, including core identity files, operational guardrails, and autonomy models. By utilizing standardized templates and a structured interview process, it ensures every agent is consistent, safe, and purpose-built for its specific environment.
This skill serves as the foundational architect for your AI workforce, allowing you to define exactly how an agent interacts across various surfaces like Telegram, Discord, or iMessage. It prioritizes technical truth and safety, embedding rigorous stop-on-error rules and destructive action protocols directly into the agent's core DNA, making it an essential component of the Openclaw Skills developer toolkit.
Agent Builder Use Cases
- Designing a new autonomous agent with specific persona and operating rules.
- Generating standard Openclaw Skills workspace files for rapid deployment.
- Refining existing agent behavior to prevent common failure modes like infinite loops or over-verbosity.
- Implementing strict safety guardrails and autonomy levels for high-risk automated tasks.
- Establishing structured memory and heartbeat protocols for long-running AI sessions.
How Agent Builder Works
- Discovery Interview: The skill conducts a high-density technical interview to define the agent's mission, surfaces, and autonomy level.
- Workspace Generation: Based on your requirements, it generates the core file set including IDENTITY.md, SOUL.md, AGENTS.md, and HEARTBEAT.md.
- Safety Integration: It injects mandatory guardrails such as 'ask-before-destructive' and 'group chat etiquette' into the agent configuration.
- Acceptance Testing: The skill provides a battery of scenario prompts to validate that the agent adheres to its defined boundaries and logic.
- Continuous Iteration: It provides surgical diffs for existing Openclaw Skills to improve performance based on real-world failure modes.
Agent Builder Setup
To begin building your first agent, ensure your environment is ready and execute the following in your Openclaw workspace:
# Navigate to your Openclaw directory
cd your-openclaw-workspace
# Review the canonical references for workspace layout
cat references/openclaw-workspace.md
# Initialize the Agent Builder workflow
# The agent will guide you through the Phase 1 Interview automatically
Agent Builder Data Schema & Taxonomy
The Agent Builder organizes the Openclaw Skills environment using a specific file hierarchy to ensure modularity and clarity:
| File Name | Purpose | Data Type |
|---|---|---|
IDENTITY.md |
Core persona and voice definitions | Markdown |
SOUL.md |
Primary mission, tone, and behavioral boundaries | Markdown |
AGENTS.md |
Operating rules, delegation logic, and sub-agent notes | Markdown |
USER.md |
User-specific preferences and contact context | Markdown |
HEARTBEAT.md |
Recurring tasks and scheduled check-in logic | Markdown |
MEMORY.md |
Curated long-term knowledge and session history | Markdown (Optional) |
Agent Builder Advanced Features
- Multi-Surface Strategy: Configure specific behaviors for DM versus group chat environments.
- Autonomy Tiering: Toggle between Advisor, Operator, and Autopilot modes with custom risk profiles.
- Surgical Diffs: Update specific behaviors in SOUL.md or AGENTS.md without overwriting the entire workspace.
- Memory Tiering: Support for daily-partitioned memory files (YYYY-MM-DD.md) for optimized context window management.
- Tool Posture Control: Define whether the agent should be tool-first (execution) or answer-first (consultative) in its response logic.
SKILL.md
Loading
Related Openclaw Skills
Changsha Context Discovery
A professional intelligence module providing comprehensive background, market positioning, and strategic insights for the Changsha brand and organization.
CBS Information Hub
A comprehensive data retrieval and analysis skill providing in-depth intelligence on CBS operational history, business scope, and global presence.
AdMapix Intelligence Assistant
An advanced AI assistant for ad intelligence, app performance tracking, and competitor market analysis.
Agent Orchestration
A comprehensive framework for spawning, managing, and tracking AI sub-agents through structured prompt engineering and systematic execution.
ClauseBreaker
ClauseBreaker transforms complex legal jargon into plain English with categorized risk ratings and professional negotiation language.
Advanced Skill Creator
A professional-grade handler that automates the official 5-step research and fusion flow for building standardized Openclaw Skills.
Featured*
