A comprehensive workspace orchestration tool for scaffolding, maintaining, and optimizing the training lifecycle of AI agents.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install openclaw-training-manager
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-training-manager 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 Training Manager?
The Training Manager is a foundational utility designed to streamline the management of an agent's personality, behavioral rules, and capabilities. It acts as the primary interface for organizing the files that define how an agent thinks and acts, ensuring that all Openclaw Skills are deployed within a robust and validated framework. By automating the creation of workspace files like SOUL.md and AGENTS.md, it transforms raw user preferences into structured instructions that guide agent logic.
Beyond simple file management, the Training Manager incorporates sophisticated security protocols. It routes all workspace modifications through sanitized scripts that apply tiered prompt injection filtering, protecting the agent from malicious overrides. This ensures that every addition to the ecosystem of Openclaw Skills remains secure, performant, and aligned with the operator's long-term goals.
Training Manager Use Cases
- Initializing a new agent environment through an interactive, conversational onboarding flow.
- Creating the structural boilerplate for custom Openclaw Skills including YAML frontmatter and requirements.
- Capturing and categorizing real-time behavioral corrections into long-term memory or personality traits.
- Auditing workspace health to identify stale files, duplicate rules, or injection risks.
- Backing up the entire training state for version control and migration.
How Training Manager Works
- The operator triggers the skill via command or during an initial setup detection phase.
- For new setups, the manager conducts a phased interview to define identity, communication style, and priorities.
- All data is processed through a translation layer that converts conversational input into optimized agent instructions.
- Files are written using specialized bash scripts that enforce rate limits and security sanitization.
- Continuous monitoring tools allow the operator to validate the workspace and consolidate fragmented updates into core files.
- The manager analyzes file sizes and complexity to ensure the agent's system prompt stays within performance limits.
Training Manager Setup
The Training Manager requires a bash environment. The workspace defaults to ~/.openclaw/workspace/ but can be customized.
# Start the interactive setup process
/training-manager setup
# Manually scaffold the workspace templates
bash scripts/scaffold.sh
# Validate the current workspace structure
bash scripts/validate.sh
Training Manager Data Schema & Taxonomy
The skill manages a structured hierarchy of Markdown files to define agent behavior:
| Component | File Path | Purpose |
|---|---|---|
| Personality | SOUL.md |
Tone, boundaries, and communication style. |
| Instructions | AGENTS.md |
Operating rules and task priorities. |
| Identity | IDENTITY.md |
Agent name, role, and versioning. |
| Context | USER.md |
Operator profile and preferences. |
| Knowledge | MEMORY.md |
Curated facts and long-term data. |
| Logs | memory/ |
Daily session logs for audit and training. |
| Skills | skills/ |
Source code and definitions for individual Openclaw Skills. |
Training Manager Advanced Features
- Multi-tiered prompt injection filtering (STRICT, NORMAL, RELAXED) tailored to file sensitivity.
- Deep workspace analysis including cross-file rule overlap detection and stale content identification.
- Automated training consolidation to merge pending updates into main instruction sets.
- Security-hardened file writers that block shell metacharacters and data exfiltration patterns.
- Metadata gating for Openclaw Skills to ensure binary and environment variable requirements are met.
SKILL.md
Loading
Related Openclaw Skills
TrustLog Guard
A privacy-first financial governance tool that monitors OpenClaw API spending, enforces budgets, and detects cost anomalies locally.
Book Debate (Multi-Lens)
A sophisticated analytical engine that uses multi-perspective debating to help users deeply understand ideas from books, thinkers, or personal notes.
Visual File Sorter
An intelligent file management agent that uses computer vision to identify, rename, and archive messy local files.
Vibe Harvester
An intelligent visual assistant that automatically scrolls through image platforms to screen and download high-quality reference photos based on your aesthetic preferences.
Jina DeepSearch
A high-performance deep search skill for AI agents to retrieve grounded information using Jina AI's advanced search models.
Wheels Router Transit Planner
A high-performance transit routing skill for AI agents providing global trip planning with deep integration for Hong Kong transportation.
Featured*
