orchestrator-worker

MIT-licensed, backend-agnostic AI agent orchestration loop in Python: orchestrator→worker→reviewer as a deterministic harness. Drive coding-agent backends through an MCP server + CLI.

Policy you install, not a framework you run: an orchestrator–worker discipline for Claude Code. One Node installer, identical on macOS, Windows, Linux, and servers. Built for codebases that get audited.

Silicon Strategy Bureau: A lightweight Multi-Agent collaboration architecture for personal knowledge workers. Zero-deployment, file-as-infrastructure, human-in-the-loop. | Mirror on Gitee: https://gitee.com/yx--jin/silicon-strategy-bureau

A study of production multi-agent patterns — planning, reflection, RAG, MCP, human-in-the-loop, guardrails, and evals — wired into one Streamlit orchestrator on the Claude Agent SDK.

Orchestrator-worker multi-agent pattern: one planner, three independent workers with identical prompts but different sampling configs (temperature/top_p). Isolates what sampling alone changes in model behavior — conservative vs balanced vs creative brainstorming.

A lightweight ai agent management system. Orchestrate one powerful AI with multiple cheap AI agents in parallel — interface contracts, file-level locks, and isolated git worktrees keep workers from colliding. Optimized for OpenCode.