SkillLift improves skill evolution efficiency using learned evaluation rubrics
SkillLift: Learning Dense Rubrics from Sparse Oracles for Efficient Skill Evolution
Artificial Intelligence
Summary
Using large language models to build reusable skills is helpful but slow because every update needs full testing. The authors found that comparing skills to see which is better is easier than scoring skills exactly, so they created SkillLift. This method learns an internal rubric to quickly judge skill improvements without always needing expensive tests. SkillLift uses a two-part learning cycle to keep the rubric accurate and guide skill updates with fewer real tests. This approach makes skill evolution more efficient and reduces the cost of improving AI agents.
What this means in practice
- •For ai system engineers: Optimize reusable skill development in AI agents by reducing the number of costly performance evaluations required.
- •For automated testing teams: Decrease computational resources spent on testing AI behaviors by leveraging learned rubrics to rank candidate skill improvements.
Authors
Haoxiang Kang, Ming Wen
Abstract
LLM-based agents increasingly rely on persistent skills, i.e., reusable procedural prompts, to adapt without weight updates. Existing skill self-evolution methods directly revise skill text based on execution feedback, but each oracle evaluation requires a full agent rollout, creating a supervision bottleneck that confines search to failure-patching updates. Our key insight is that ranking is a smoother supervision target than absolute outcome regression: identifying which skill is better requires fewer oracle evaluations than predicting exact scores. Building on this insight, we propose SkillLift, which decouples skill search from oracle cost by learning an oracle-aligned rubric as a structured evaluation space. We formalize this as a bilevel optimization problem solved via alternating optimization: an inner loop uses the frozen rubric as a cheap surrogate to guide skill revision at no oracle cost, while an outer loop invokes a small number of oracle rollouts to re-align the rubric via rank correlation, amortizing oracle cost and stabilizing text-space updates. Experiments on complex agent task benchmarks show that our method outperforms existing auto-skill methods with 40--70\% less token cost compared to frontier evolving methods. Codes are available at https://github.com/WalteR-MittY-pro/SkillLift.