agent-registry
matrixy 作
AI & LLMs概要
MANDATORY agent discovery system for token-efficient agent.
説明
このスキルでできること
Agent Registry is a lazy-loading discovery system for Claude Code agents. Instead of loading all agents upfront, it indexes them and loads only what's needed for a given task. Reduces context window usage by roughly 95% compared to loading all agents at startup.
Avoids the ~16k token overhead of loading all agents upfront by indexing them and fetching only the ones relevant to each task.
使いどころ
- Finding the right specialized agent for a code review task
- Listing all available agents when starting a new project
- Loading a security auditor agent on demand for vulnerability checks
- Migrating existing ~/.claude/agents/ files into the registry
- Running complex multi-step workflows without hitting token limits
ワークフロー例
User asks: Can you review my authentication code for security issues?
- 1
Run search.js with keywords extracted from the request: bun bin/search.js "code review security authentication"
- 2
Review relevance scores in results to identify best-matching agent
- 3
Load the top-scoring agent's full instructions: bun bin/get.js security-auditor
- 4
Follow the loaded agent's instructions to perform the security review
Security audit of the authentication code using the loaded agent's specialized instructions
要件
- Bun runtime (ships with Claude Code, no separate install needed)
インストール
clawhub install matrixy/agent-registryCLIコマンドを実行するか、Claude Code、Cursor、Codexなどお使いのコーディングエージェントにプロンプトを貼り付けてください。