SkillLift: Learning Dense Rubrics from Sparse Oracles for Efficient Skill Evolution

• URL: http://arxiv.org/abs/2609.15396v1
• Date: Mon, 14 Sep 2026 11:30:23 GMT
• Title: SkillLift: Learning Dense Rubrics from Sparse Oracles for Efficient Skill Evolution
• Abstract summary: LLM-based agents increasingly rely on persistent skills, i.e., reusable procedural prompts, to adapt without weight updates.<n>Existing skill self-evolution methods directly revise skill text based on execution feedback, but each oracle evaluation requires a full agent rollout.<n>We propose SkillLift, which decouples skill search from oracle cost by learning an oracle-aligned rubric as a structured evaluation space.<n> 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.
• Score: 2.429897021787139
• License: http://creativecommons.org/licenses/by/4.0/
• 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.

Related papers

• Write, Execute, Refine: From Skill Followers to Skill Optimizers via Reinforcement Learning from Execution Feedback [52.78692582507693]
  Expert-written natural language skills can improve tool-using agents, yet agent-authored skills perform 8-11 points worse than using no skill.<n>We study how to organize execution experience from intermediate skills into training states for an executor.
  arXiv  Detail & Related papers  (2026-08-18T09:52:48Z)
• Escaping the Self-Repair Trap: Improving Test Oracle Generation via Dual-Context Awareness [4.39623958563083]
  DCAware is a non-iterative framework that prioritizes high signal-to-noise contextual grounding over multi-round repair.<n>We show that DCAware consistently improves fault-revealing effectiveness while maintaining high execution success.<n>Our results suggest that improving contextual quality is more effective than adding iterative repair complexity.
  arXiv  Detail & Related papers  (2026-08-06T11:43:11Z)
• REVES: REvision and VErification--Augmented Training for Test-Time Scaling [53.197756110943395]
  We propose a two-stage iterative framework that alternates between online data/prompt augmentation and policy optimization.<n>We observe gains of +6.5 points over the RL baseline and +4.0 points over standard multi-turn training.
  arXiv  Detail & Related papers  (2026-06-17T10:37:23Z)
• Scaling Laws for Agent Harnesses via Effective Feedback Compute [53.68149869349268]
  emphEffective Feedback Compute (EFC) is a trace-level scaling coordinate that credits feedback only when it is informative, valid, non-redundant, and retained for subsequent decisions.<n>EFC-based coordinates consistently predict failure rates better than raw-compute baselines.
  arXiv  Detail & Related papers  (2026-05-28T09:45:47Z)
• Reducing Oracle Feedback with Vision-Language Embeddings for Preference-Based RL [18.45416397753188]
  ROVED is a hybrid framework that combines VLE-based supervision with targeted oracle feedback.<n>Our method uses the VLE to generate segment-level preferences and defers to an oracle only for samples with high uncertainty.<n>Across multiple robotic manipulation tasks, ROVED matches or surpasses prior preference-based methods while reducing oracle queries by up to 80%.
  arXiv  Detail & Related papers  (2026-03-30T05:33:55Z)
• SQL-ASTRA: Alleviating Sparse Feedback in Agentic SQL via Column-Set Matching and Trajectory Aggregation [51.696027315573296]
  We propose a framework featuring a universal two-tiered reward mechanism designed to provide effective trajectory-level evaluation and dense step-level signals.<n>In traditional paradigms, rewards are determined solely by the final-turn feedback, which ignores the intermediate process and leads to ambiguous credit evaluation.<n>We show that our approach outperforms SOTA Arctic-Text2-R1-7B on BIRD 2.0 using identical models, propelling Text-to-reward convergence toward a robust multi-turn agent paradigm.
  arXiv  Detail & Related papers  (2026-03-17T06:26:36Z)
• Fast or Better? Balancing Accuracy and Cost in Retrieval-Augmented Generation with Flexible User Control [52.405085773954596]
  Retrieval-Augmented Generation has emerged as a powerful approach to mitigate large language model hallucinations.<n>Existing RAG frameworks often apply retrieval indiscriminately,leading to inefficiencies-over-retrieving.<n>We introduce a novel user-controllable RAG framework that enables dynamic adjustment of the accuracy-cost trade-off.
  arXiv  Detail & Related papers  (2025-02-17T18:56:20Z)
• Necessary and Sufficient Oracles: Toward a Computational Taxonomy For Reinforcement Learning [28.184175745050474]
  We study the impact of the choice of supervised learning oracle on the computational complexity of reinforcement learning algorithms.<n>First, we identify two-context regression as a minimal oracle in the standard episodic access model.<n>Second, we identify one-context regression as a near-minimal oracle in the stronger reset access model.<n>Third, we broaden our focus to Low-Rank MDPs, where we give cryptographic evidence that the analogous oracle from the Block MDP setting is insufficient.
  arXiv  Detail & Related papers  (2025-02-12T18:47:13Z)
• Improve the efficiency of deep reinforcement learning through semantic exploration guided by natural language [10.47685316733524]
  We propose a novel method for interacting with the oracle in a selective and efficient way, using a retrieval-based approach. We use a neural network to encode the current state of the agent and the oracle, and retrieve the most relevant question from the corpus to ask the oracle. We show that our method can significantly improve the efficiency of RL by reducing the number of interactions needed to reach a certain level of performance.
  arXiv  Detail & Related papers  (2023-09-21T03:25:35Z)
• Lero: A Learning-to-Rank Query Optimizer [49.841082217997354]
  We introduce a learning to rank query, called Lero, which builds on top of the native query and continuously learns to improve query optimization. Rather than building a learned from scratch, Lero is designed to leverage decades of wisdom of databases and improve the native. Lero achieves near optimal performance on several benchmarks.
  arXiv  Detail & Related papers  (2023-02-14T07:31:11Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.