Clawtool

•
4
•

Provides a canonical tool layer for AI coding agents, offering enhanced core utilities, multi-agent supervision, and intelligent tool discovery.

Clawtool functions as a centralized supervisor and tool layer for AI coding agents, aggregating native capabilities and third-party sources into a cohesive interface. It replaces standard operating system commands with timeout-safe, format-aware, and structured-output equivalents, ensuring deterministic and secure agent operations across various environments. The tool features a multi-agent dispatch system utilizing the Bidirectional Inter-Agent Messaging (BIAM) protocol for asynchronous task delegation, and a search-first discovery mechanism (ToolSearch) that allows agents to find and bind to relevant tools based on natural-language intent. Additionally, it integrates as a Claude Code marketplace plugin, streamlines subagent scaffolding, and offers sandboxing capabilities for isolated execution.

주요 기능

010 GitHub stars
02Search-first tool discovery using bleve BM25 ranking for natural-language intent-based tool binding.
03Multi-agent dispatch with BIAM protocol for asynchronous delegation between Claude, Codex, Gemini, or Opencode agents.
04Canonical core tools (Bash, Read, Edit, Write, Grep, Glob) with structured JSON output and timeout safety.
05First-class Claude Code marketplace plugin for zero-ceremony registration and skill injection.
06Configurable sandbox profiles (bwrap, sandbox-exec, docker) for isolated and secure agent operations.

사용 사례

01Expanding an AI agent's capabilities with robust, reliable, and secure core tools for file manipulation, search, and command execution.
02Scaffolding and managing new subagent personas, skills, or MCP servers to customize and extend agent functionalities.
03Delegating complex or long-running tasks between different AI agents without blocking conversations.