mlsanityData to 14 Oct 2026

SkillOpt: Executive Strategy for Self-Evolving Agent Skills

arXiv 2605.23904 ↗First posted 22 May 2026cs.AI cs.CLcited by 148

On the map: LLM agents and tool use → Agent skill learning and evolution

also in 1 more
Abstract (arXiv)

Agent skills today are hand-crafted, generated one-shot, or evolved through loosely controlled self-revision, none of which behaves like a deep-learning optimizer for the skill, and none of which reliably improves over its starting point under feedback. We argue the skill should instead be trained as the external state of a frozen agent, with the same discipline that makes weight-space optimization reproducible. SkillOpt is, to our knowledge, the first systematic controllable text-space optimizer for agent skills: a separate optimizer model turns scored rollouts into bounded add/delete/replace edits on a single skill document, and an edit is accepted only when it strictly improves a held-out validation score. A textual learning-rate budget, rejected-edit buffer, and epoch-wise slow/meta update make skill training stable while adding zero inference-time model calls at deployment. Across six benchmarks, seven target models, and three execution harnesses (direct chat, Codex, Claude Code), SkillOpt is best or tied on all 52 evaluated (model, benchmark, harness) cells and beats every per-cell competitor among human, one-shot LLM, Trace2Skill, TextGrad, GEPA, and EvoSkill skills. On GPT-5.5 it lifts the average no-skill accuracy by +23.5 points in direct chat, by +24.8 inside the Codex agentic loop, and by +19.1 inside Claude Code. Transfer experiments further show that optimized skill artifacts retain value when moved across model scales, between Codex and Claude Code execution environments, and to a nearby math benchmark without further optimization. Code: https://aka.ms/skillopt

Key facts. Cited by 148 papers in the map (data through 14 Oct 2026). Placed in Agent skill learning and evolution, one of 16 directions in LLM agents and tool use. Builds on 32 of its 48 references that are in the map. First posted on arXiv 22 May 2026.

Builds on

32 of 48 references.

Closest ancestors · 10

Same direction, newest first

  1. 2026SkillForge: Forging Domain-Specific, Self-Evolving Agent Skills in Cloud Technical Support9 Apr
  2. 2026SkillClaw: Let Skills Evolve Collectively with Agentic Evolver9 Apr
  3. 2026SkillX: Automatically Constructing Skill Knowledge Bases for Agents6 Apr
  4. 2026CoEvoSkills: Self-Evolving Agent Skills via Co-Evolutionary Verification2 Apr
  5. 2026Trace2Skill: Distill Trajectory-Local Lessons into Transferable Agent Skills26 Mar
  6. 2026EvoSkill: Automated Skill Discovery for Multi-Agent Systems3 Mar
  7. 2026SoK: Agentic Skills -- Beyond Tool Use in LLM Agents24 Feb
  8. 2026SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks13 Feb
  9. 2026SkillRL: Evolving Agents via Recursive Skill-Augmented Reinforcement Learning9 Feb
  10. 2025Reinforcement Learning for Self-Improving Agent with Skill Library18 Dec

Other references · 22

  1. 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,695
  2. 2023Reflexion: Language Agents with Verbal Reinforcement Learningcited by 4,276
  3. 2023Toolformer: Language Models Can Teach Themselves to Use Toolscited by 4,182
  4. 2023Self-Refine: Iterative Refinement with Self-Feedbackcited by 3,442
  5. 2023Voyager: An Open-Ended Embodied Agent with Large Language Modelscited by 2,191
  6. 2024SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineeringcited by 1,737
  7. 2020DocVQA: A Dataset for VQA on Document Imagescited by 1,317
  8. 2020ALFWorld: Aligning Text and Embodied Environments for Interactive Learningcited by 1,205
  9. 2024OlympiadBench: A Challenging Benchmark for Promoting AGI with Olympiad-Level Bilingual Multimodal Scientific Problemscited by 1,136
  10. 2023Large Language Models as Optimizerscited by 864
  11. 2023DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelinescited by 832
  12. 2025GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learningcited by 542
10 more
  1. 2024TextGrad: Automatic "Differentiation" via Textcited by 273
  2. 2017SearchQA: A New Q&A Dataset Augmented with Context from a Search Enginecited by 225
  3. 2025EvolveR: Self-Evolving LLM Agents through an Experience-Driven Lifecyclecited by 137
  4. 2026AutoSkill: Experience-Driven Lifelong Learning via Skill Self-Evolutioncited by 115
  5. 2024SpreadsheetBench: Towards Challenging Real World Spreadsheet Manipulationcited by 111
  6. 2025Memp: Exploring Agent Procedural Memorycited by 103
  7. 2026SKILLFOUNDRY: Building Self-Evolving Agent Skill Libraries from Heterogeneous Scientific Resourcescited by 41
  8. 2026OfficeQA Pro: An Enterprise Benchmark for End-to-End Grounded Reasoningcited by 39
  9. 2025EvoTest: Evolutionary Test-Time Learning for Self-Improving Agentic Systemscited by 37
  10. 2026ABSTRAL: Automatic Design of Multi-Agent Systems Through Iterative Refinement and Topology Optimizationcited by 6

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 148.

148 papers.

Sort:

By area:

  1. 2026SkillOpt-Lite: Better and Faster Agent Self-evolution via One Line of Vibecited by 20
  2. 2026Self-Improvements in Modern Agentic Systems: A Surveycited by 19
  3. 2026DarwinX: Evolving Agent Harnesses Through Natural Selectioncited by 13
  4. 2026EnvHarness: Awakening Static Worlds for Agent Learningcited by 13
  5. 2026RRSI: Regularized Recursive Self-Improvement of Agent Harnessescited by 13
  6. 2026WikiSkill: Compiling Agent Experience into Persistent Knowledge for Skill Evolutioncited by 12
  7. 2026PAST-Bench: Benchmarking the Foundations of Recursive Self-Improvement in Personal Agentscited by 9
  8. 2026HarnessEvolve: Learning from Reference Trajectories for Reliable Agent Self-Evolutioncited by 9
  9. 2026Hypothesis-Driven Skill Optimization for LLM Agentscited by 8
  10. 2026SkillCoach: Self-Evolving Rubrics for Evaluating and Enhancing Agentic Skill-Usecited by 8
  11. 2026RSIAgent: Autonomous Exploration for Recursive Self-improvement in New Environmentscited by 8
  12. 2026VideoWeaver: Evaluating and Evolving Skills for Agentic Long Video Generationcited by 7
38 more
  1. 2026A Framework for Evaluating Agentic Skills at Scalecited by 7
  2. 2026Rethinking Self-Evolving Agent Skills: Feedback Dynamics over Multiple Roundscited by 7
  3. 2026Hierarchical Self-Improvement: A Framework for Task-Specific Evolvable Agent Harnessescited by 7
  4. 2026ScienceBuddy: Recursive-in-Recursive Self-Improvement for Interactive Scientific Agentscited by 7
  5. 2026SkillJuror: Measuring How Agent Skill Organization Changes Runtime Behaviorcited by 6
  6. 2026Dynamic Agent Skills: A Lifecycle Survey and Taxonomy of Evolving Skill Librariescited by 6
  7. 2026SkillProx: Self-Evolving Agent Skills via Proximal Textual Gradient Descentcited by 6
  8. 2026SHE: Trajectory-driven Safety Harness Evolution for LLM Agentscited by 6
  9. 2026ACE-Brain-0.5: A Unified Embodied Foundational Model for Physical Agentic AIcited by 5
  10. 2026SkillRise: Agentic Reinforcement Learning for Cross-Task Skill Evolutioncited by 5
  11. 2026DIVE: Unlocking Self-Improvement in Frozen Language Models Through Diversity-Driven Skill Evolutioncited by 5
  12. 2026Agent Skills Matter: Inferring Proprietary Skills from Execution Trajectoriescited by 4
  13. 2026AgentStream: How Well Do Self-Evolving LLM Agents Perform Under Streaming Tasks?cited by 4
  14. 2026KV-Skill: Forging Expertise in the Model's Native Languagecited by 4
  15. 2026SkillZip: Evaluation-Free Skill Compression for Self-Evolving Agents by Discovering Reusable Structurecited by 4
  16. 2026SoftSkill: Behavioral Compression for Contextual Adaptationcited by 3
  17. 2026Parametric Skillscited by 3
  18. 2026SKILL-KD: Contrastive Skill Distillation for LLM Agentscited by 3
  19. 2026Rethinking Self-Evolving Agents: Do We Still Need Prescribed Optimization Pipelines?cited by 3
  20. 2026Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agentscited by 3
  21. 2026Agentic ESOpt: Fine-Tuning Long-Horizon LLM Agents with Minimal GPU Requirementscited by 3
  22. 2026Graph Engineering in the Era of LLM Agents: From Individual Intelligence to System Intelligencecited by 3
  23. 2026SkillAdam: Stable and Efficient Skill Evolution for Agentscited by 3
  24. 2026Unsupervised Skill Discovery for Agentic Data Analysiscited by 2
  25. 2026Workflow-Localized Mechanism Learning: Attribution-Guided Repair and Knowledge Reuse for Structured Agent Skillscited by 2
  26. 2026DecoEvo: Score-Decoupled Co-Evolution of Solver and Rubric-Generator Skills in Text Spacecited by 2
  27. 2026What Is a Skill Worth? Structure-Aware Shapley Valuation of Agent Skillscited by 2
  28. 2026Self-Evolving Code-with-Image Reasoningcited by 2
  29. 2026DataFoundry: Evolving Data Preparators via Recursive Self-Improvementcited by 2
  30. 2026EvoSCM: Scientific Belief Revision Through Causal Model Evolution and Experimentationcited by 2
  31. 2026SkillGLoW: Procedural-Family Skill Consolidation for Self-Improving Agents on Long-Horizon Task Streamscited by 2
  32. 2026Skill-3D: Evolving Scene-Aware Skills for Agentic 3D Spatial Reasoningcited by 1
  33. 2026Marginal Advantage Accumulation for Memory-Driven Agent Self-Evolutioncited by 1
  34. 2026An Empirical Study of Downstream Adaptation for Agent Skillscited by 1
  35. 2026Learning on the Job: Continual Learning from Deployment Feedback for Frozen-Weights Agentscited by 1
  36. 2026Rethinking Self-Evolution: A Constrained Exploration-Exploitation Process for Mitigating Skill Overfittingcited by 1
  37. 2026MIEScore: Human-Aligned Evaluation for Multi-Source Image Editingcited by 1
  38. 2026ADIAS: Automated Design of Interactive Agentic Systemscited by 1

50 most cited of 148 All 148 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2026SkillComposer: Learning to Evolve Agent Skills for Specification and Generalization
  2. 2026MetaSkill-Evolve: Recursive Self-Improvement of LLM Agents via Two-Timescale Meta-Skill Evolution
  3. 2026Skills-Coach: A Self-Evolving Skill Optimizer via Training-Free GRPO
  4. 2026SkillHone: A Harness for Continual Agent Skill Evolution Through Persistent Decision History
  5. 2026SkillMaster: Toward Autonomous Skill Mastery in LLM Agents
  6. 2026SkillFocus: Evolving Agent Skills via Capability Decomposition
  7. 2026SkillGym: Internalizing Human Skills into LLMs for Real-World Problem Solving
  8. 2026SkillAdaptor: Self-Adapting Skills for LLM Agents from Trajectories