012 GitHub stars
02Cost-optimized delegation of coding tasks from Claude Code to local or flat-rate LLMs.
03MCP server exposing `run_task`, `reply_task`, and `check_task` for bidirectional sub-agent communication.
04FastAPI gateway for LLM routing, failover, caching, and concurrent request management.
05Supports multiple executor LLMs including Ollama (local) and z.ai (flat-rate).
06Bundled reference skills (`/feature-develop-hermit`, `/code-apply-hermit`, etc.) demonstrating the delegation pattern.