A comprehensive framework for onboarding, supervising, and evolving multi-agent teams to ensure mission alignment and behavioral consistency.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-training
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-training 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 Training System?
The Agent Training skill functions as a Main Agent or instructor role, designed to manage the entire lifecycle of sub-agents within a collaborative environment. By utilizing Openclaw Skills, it provides a structured methodology for ensuring that every member of an AI team shares the same mission, understands user requirements, and adheres to unified behavioral standards. This system bridges the gap between individual agent performance and collective team intelligence.
At its core, the skill maintains a centralized training manual and supervises the configuration of each agent's identity, persona, and memory. It is essential for developers and power users who need to scale their AI operations while maintaining strict control over agent behavior, privacy protocols, and evolutionary progress.
Agent Training System Use Cases
- Creating and onboarding new sub-agents with standardized identity and soul configurations.
- Maintaining a centralized AGENT_TRAINING.md to align team-wide missions and success criteria.
- Supervising agent status and performing regular configuration audits to prevent drift.
- Implementing behavioral guardrails such as quiet hours and interaction frequency limits.
- Driving team evolution through daily retrospectives and weekly performance reviews.
How Agent Training System Works
- The instructor agent initializes the AGENT_TRAINING.md manual containing the team structure and code of conduct.
- For each new agent, a specific directory is created containing IDENTITY.md, SOUL.md, MEMORY.md, and HEARTBEAT.md files.
- The user profile (USER.md) is synchronized to the sub-agent's local environment to provide necessary context.
- The sub-agent undergoes a training phase where it reads the manual and records its understanding and responsibilities in its MEMORY.md.
- The instructor uses supervision commands to monitor the heartbeat and execution status of the entire team.
- Feedback loops are established via daily reviews and weekly evolution cycles to update the Openclaw Skills configurations.
Agent Training System Setup
To deploy the training system within your Openclaw Skills workspace, prepare the directory structure and initialize the training manual:
# Create the agent directory structure
mkdir -p agents/{agent_id}
# Initialize the core training manual
touch AGENT_TRAINING.md
# Copy the user profile to the new agent
cp USER.md agents/{agent_id}/USER.md
After setting up the files, trigger the training process by using the keyword 'team training' to have the instructor agent begin the onboarding sequence.
Agent Training System Data Schema & Taxonomy
The skill utilizes a distributed file-based schema to manage agent state and team logic within the Openclaw Skills ecosystem:
| File Name | Description | Metadata Captured |
|---|---|---|
AGENT_TRAINING.md |
Team Manual | Missions, Structure, Rules, Collaboration mechanisms |
IDENTITY.md |
Role Definition | Agent ID, primary responsibilities, and constraints |
SOUL.md |
Persona Config | Personality traits, tone of voice, and emotional logic |
MEMORY.md |
Experience Log | Training records, user feedback, and historical context |
HEARTBEAT.md |
Task Scheduler | Active status and recurring automated tasks |
Agent Training System Advanced Features
- Multi-agent supervision with commands like
review all agentsandagent statusfor comprehensive team oversight. - Intelligent silence protocols that prevent agents from interacting during specified hours (23:00-8:00) or after explicit user requests.
- Automated evolution mechanism including weekly configuration audits and information cleanup.
- Standardized confirmation templates for sub-agents to verify their understanding of team missions and behavioral rules.
- Privacy-first data handling that prevents sensitive memory information from leaking into group communications.
SKILL.md
Loading
Related Openclaw Skills
Writing Skills
A framework that applies Test-Driven Development principles to documentation, ensuring AI agents follow patterns and techniques reliably.
Writing Plans
A systematic blueprinting tool for AI agents to generate granular, TDD-focused implementation plans before writing code.
Verification Before Completion
A rigorous technical protocol that mandates fresh evidence from tests and builds before any work is claimed as finished.
Using Superpowers: AI Skill Orchestration
A mandatory meta-skill that enforces the invocation of relevant specialized skills before an AI agent takes any action or answers any question.
Feishu Agent Integration
A comprehensive framework for creating custom AI agents and integrating them into Feishu messaging environments using Openclaw Skills.
WeChat Publisher
Automate the publication of articles and management of drafts on WeChat Official Accounts using these Openclaw Skills.
Featured*
