← All papers
First page of SkillEvoLean: Mutation-enhanced skill evolution for Lean provers

SkillEvoLean: Mutation-enhanced skill evolution for Lean provers

Kuo Zhou, Zixiong Yang, Lu Zhang

cs.LG Oct 1, 2026 · v3
Evolves LLM agent skills for Lean theorem proving, using the Lean verifier and language server for feedback, evaluated on Lean benchmarks.
Skill evolution offers a promising way to improve large language model agents without updating their parameters, but its use in formal theorem proving remains underexplored. Existing methods mainly target natural-language reasoning, improving skills by analyzing successful and failed trajectories and incrementally revising solving strategies. Although the Lean verifier provides reliable execution feedback, when all sampled trajectories fail, existing skill evolution methods lack successful trajectories from which to infer effective update directions. Furthermore, these methods also focus mainly on the root instruction file, thus underexploring the evolution of reference knowledge including mathematical concepts and proving techniques. To address these limitations, we propose a mutation-enhanced skill self-evolution framework for building skill-augmented Lean provers. The framework jointly evolves a high-level solving policy and its reference knowledge through progressive and mutation-based updates. Progressive evolution derives local improvements from successful and failed trajectories, while mutation is triggered when no complete proof can be generated, sampling mathematical concepts to produce and select new skill candidates under verifier feedback. We evaluate our method on MiniF2F, PutnamBench, the 2025 International Mathematical Olympiad (IMO 2025), and the 2026 USA Mathematical Olympiad (USAMO 2026). Under the same backbone model, trajectorysampling budget, and test-time compute, our method achieves proof success rates of 100.0%, 90.6%, 4/6, and 4/6, respectively, with GPT-5.5, outperforming the baseline methods. Further analysis shows that concept-guided mutation outperforms random-text-guided mutation by 6.9 and 8.2 percentage points on MiniF2F and PutnamBench, respectively, while solving one additional problem on both IMO 2025 and USAMO 2026.

Skill evolution improves LLM agents without parameter updates, but it has seen little use in formal theorem proving. Existing methods stall when every sampled proof trajectory fails, because no successful trajectory exists to suggest an update. They also rarely evolve reference knowledge such as mathematical concepts and proving techniques.

SkillEvoLean jointly evolves a high-level solving policy (SKILL.md) and its reference knowledge. Progressive evolution uses LLM operators to compare successful and failed trajectories and then locate, suggest and rewrite skill edits. When no complete proof is found, concept-guided mutation samples new mathematical concepts to perturb the skill. The Lean verifier selects among the resulting candidates. Agents interact with the Lean language server through a Claude Code harness.

With GPT-5.5, it reaches 100.0% on MiniF2F, 90.6% on PutnamBench, 4/6 on IMO 2025 and 4/6 on USAMO 2026, outperforming the No Skill, Human Skill, LLM Skill, Trace2Skill and SkillOpt baselines. Concept-guided mutation beats random-text mutation by 6.9 points on MiniF2F and 8.2 points on PutnamBench, and solves one more problem on each olympiad set.

Figure 3: Progressive evolution converges without mutation. Without mutation-based exploration, progressive updates gradually saturate.
BenchmarkSkillEvoLean
MiniF2F100.0%
PutnamBench90.6%
IMO 20254/6
USAMO 20264/6
Proof success with GPT-5.5