Maestro: Agentic Formal Proof Development by Coordination

Key points

• Lean is increasingly used to verify mathematical proofs and software produced by LLM agents, but writing formal proofs remains hard.1
• Existing prover systems take a divide-and-conquer approach that recursively decomposes hard proof obligations into easier lemmas and splits again any lemma that resists proof.1
• We evaluate Maestro with four frontier LLMs on seven mathematics and three software verification benchmarks.1
• With Claude Opus 5, it discovers its own proofs of 34 of the 48 theorems in the June 2026 release of arXivLean, against 19 for the native agent.1

Sources (1)

• [1]Maestro: Agentic Formal Proof Development by Coordination
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 07:21 AM

  We introduce Maestro, a Lean prover harness that handles failures through global coordination.

  Lean is increasingly used to verify mathematical proofs and software produced by LLM agents, but writing formal proofs remains hard.

Extractive summary: sentences quoted from the sources.