agent-registry

What This Skill Does

The Agent Registry is a mandatory, high-efficiency system designed to revolutionize how OpenClaw interacts with specialized AI agents. By replacing the manual, resource-heavy practice of loading all agents from local storage, this skill introduces a lazy-loading architecture. It effectively reduces context window usage by approximately 95%. When you initiate a task, the registry allows the agent to search for, identify, and load only the specific instructions required for the immediate job. This ensures that Claude's processing power and memory are reserved for task execution rather than maintaining massive, unused instruction sets. It acts as an intelligent directory, enabling seamless scaling even if your environment contains hundreds of unique agent configurations.

Installation

You can install this skill easily using the Skills CLI. For the most streamlined experience, run npx skills add MaTriXy/Agent-Registry@agent-registry. Alternatively, for a traditional manual install, you can run ./install.sh in the repository directory, or append --project for local-only scoping. If you want the most visually intuitive management experience, use the --install-deps flag to pull in @clack/prompts, which provides a beautiful, interactive CLI interface with color-coded token usage indicators. After installation, execute bun bin/init.js to migrate your existing agent library into the new, optimized registry structure.

Use Cases

This skill is essential when you have a sprawling library of specialized assistants and want to avoid the notorious '16k token limit' warning. Use it when you are starting complex workflows that require specialized domain knowledge (such as security auditing or specialized code refactoring), when you need to quickly look up which tools you have available for a specific task, or when you are performing multi-step automation where different agents should be swapped in and out dynamically.

Example Prompts

1. "I need to run a security audit on the user authentication module; which of my agents would be best for this?"
2. "Search for all agents related to SQL database optimization and give me a brief overview of their capabilities."
3. "Load the documentation-generator agent so we can begin updating the project READMEs."

Tips & Limitations

Always follow the 'Search First' pattern: execute a search before attempting to load an agent. This ensures you are pulling the most relevant version and keeping your context clean. Note that the registry relies on file-system access to your agent directory; ensure that the registry is correctly initialized using the migration script to prevent pathing issues. While the search tool handles large lists, very high volumes of agents (over 500) may benefit from using the search-paged.js command to maintain optimal performance.