Agent0: The Self-Evolving Loop

An interactive analogy based on the paper "Agent0: Unleashing Self-Evolving Agents from Zero Data".
The Riddle Master (Curriculum Agent) creates puzzles to challenge The Detective (Executor Agent).
They grow smarter together without any outside help.
🧙‍♂️ Riddle Master (Curriculum Agent) Complexity Level 🕵️‍♀️ Detective (Executor Agent) Skill Level Generating... Solving...

🧙‍♂️ The Riddle Master

Goal: Create riddles that are just right.

Rewards:

  • Confusion: If the Detective is unsure (waffles between answers).
  • Gadget Use: If the Detective MUST use tools to solve it.

"If it's too easy, I learn nothing. If it's impossible, I learn nothing. I need the sweet spot."

🕵️‍♀️ The Detective

Goal: Solve the riddles correctly.

Method:

  • Reasoning: Thinking through the problem step-by-step.
  • Tools: Using the "Gadget Belt" (Python Interpreter) for hard math/logic.
  • Self-Consistency: Asking "inner voices" to vote on the best answer.

> System initialized. > Ready to start evolution cycle.