The problem
LangGraph’s ideas — nodes, shared state, conditional edges, recursion limits, human approval — are easy to read about and hard to picture.
How it works
- 01Read the goal
- 02Connect the arrows
- 03Run the tests
- 04Watch each path
- 05Earn stars
What was hard
- A tiny graph engine in the LangGraph style: each node returns a state update and a route; fixed and conditional edges pick what runs next.
- Every level is checked with test inputs, so a wrong arrow fails with a plain sentence — “paid ₹50,000 with nobody checking”.
- Loops are stopped at a recursion limit, as LangGraph does, so the player sees why agents need a way out.
The result
Four levels covering the four patterns most agents are built from.