agent-registry
A lazy loading system for Claude Code that provides on-demand Agent loading, significantly saving context tokens.
Install & Use
Copy this prompt and send it to your AI assistant (Claude / Cursor / TRAE / Codex / WorkBuddy etc.) to auto-install:
Help me install this AI Skill: agent-registry. It is used for: A lazy loading system for Claude Code that provides on-demand Agent loading, significantly saving context tokens. Full Skill content: https://321skill.com/skills/agent-registry/raw/index.md Read that page and install it.
The prompt includes a link to the full Skill content. You can also view the full content.
As the number of installed Agents increases, Claude Code by default loads all Agents in every conversation, consuming a large amount of valuable context tokens with unused Agents. Agent Registry changes the Agent loading method from "preload all" to "on-demand loading" by building an index and implementing a lazy loading mechanism, thereby reducing Agent-related token consumption by 70-90%.
When you use it, you simply ask questions in Claude Code as usual. Agent Registry automatically analyzes your query intent via a built-in BM25 search engine and matches the most relevant Agents in real-time. When a high-confidence Agent is matched, the system injects its instructions into the context. The entire process is transparent to the user, requiring no manual search or intervention.
This Skill is ideal for developers, data analysts, or content creators who have installed dozens or even hundreds of Agents in Claude Code but only use a few in each conversation. It significantly increases available context space, giving you more "working memory" for complex coding, analysis, or writing tasks.
Consider using it when the number of Agents exceeds 20. After installation, your existing Agent usage habits remain unchanged, as the system optimizes automatically. Please note that its auto-discovery feature relies on semantic analysis of user queries; for extremely vague or brief instructions, it may not trigger the most precise Agent matching.
Key Features
The core differentiator is its "fully automatic, seamless" lazy loading mechanism: via hooks, it performs real-time semantic search on each user query and injects matched Agents, requiring no manual search or management from the user. This saves a significant number of tokens while maintaining a seamless user experience.
Limitations
Not suitable for users who have only installed a few Agents or wish to have full manual control over when each Agent is loaded.
FAQ
Will installing it affect my existing Agent usage?
No. Your usage pattern remains completely unchanged. Agent Registry works automatically in the background, loading Agents on-demand to save tokens.
How much context space can it save?
According to repository data, in a scenario with 140 installed Agents, it can save approximately 83% of Agent-related tokens, freeing up significant space for actual conversation and code.
Installation guide for AI assistants
If your AI coding assistant (Claude Code, Cursor, TRAE etc.) can see this page, send it this message to auto-install:
Visit https://321skill.com/skills/agent-registry/raw/index.md to read the original Skill definition (Markdown format) for agent-registry, and install it according to the instructions.
Raw Markdown URL for AI: /skills/agent-registry/raw/index.md