muhammadqasimkalhoro94-blip/claude-zeroclaw-agentics

This repository presents a tool called Claude-Zeroclaw, described as a framework for running multiple Claude AI sub-agents across different code repositories and tool servers simultaneously while sharing a common context cache. The README frames the core problem as Claude Code operating in a single bounded session per invocation, and positions Zeroclaw as a way to coordinate several concurrent Claude instances without duplicating context between sessions. The README describes five main components: a Zeroclaw Router that directs requests to the appropriate sub-agent, a Context Cache Engine described as reducing token usage by approximately 60 percent, a Sub-Agent Manager that spawns and coordinates multiple Claude instances, an MCP Proxy that combines multiple MCP server connections into a single stream, and a Desktop Bridge in beta that links Claude Desktop with Claude Code running in a terminal. A diagram in the README shows a central Claude instance routing tasks to four sub-agents covering code analysis, MCP integration, desktop control, and SDK interaction. A YAML configuration example in the README shows how sub-agents are defined with model assignments, per-agent instructions, and designated MCP server connections. A code review profile distributes static analysis, dependency auditing, and documentation updates across three separate agents, each using a different Claude model. Multi-repo sweep mode and desktop bridge mode are demonstrated in command-line examples. Installation is via pip as claude-zeroclaw or from source. Prerequisites include Python 3.10 or later, the Claude Code CLI, and Node.js 18 or later for MCP proxy support. The README lists Linux and macOS as fully supported, Windows 11 as beta, and WSL2 as fully supported. Android via Termux is listed as experimental. The README is 13,323 characters and lists no programming language.