HanLi-Chainsea/han-agents ? reverse-engineered prompt

Build me a Python tool that plugs into coding assistants like Claude Code, Cursor, and similar apps, and gives them a more reliable way to handle big coding tasks without losing context.

I want it to feel zero setup. After cloning, the first run should create its local database, register whatever hooks are needed, and install code parsers automatically. The core idea is a brain inspired multi agent workflow with persistent memory, a live map of the codebase, a task system with parent and child tasks, and checkpoints so long jobs can pause and resume across chats.

It should also guide the agent with reusable skill definitions, detect when the real code drifts away from those definitions, and support recipe style shortcuts that can generate a full task tree from one request. Add an automatic executor, critic, memory loop, plus local tracing and evals.

Please include solid unit and integration test commands that actually verify results, check branch coverage, and reject fake integration tests that only use mocks. Auto detect the project stack if possible. Look up current docs online if you need to.