mlsanityData to 7 Oct 2026

SkillReducer: Optimizing LLM Agent Skills for Token Efficiency

arXiv 2603.29919 ↗First posted 31 Mar 2026cs.SEcited by 29

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

Abstract (arXiv)

LLM-based coding agents rely on skills, pre-packaged instruction sets that extend agent capabilities, yet every token of skill content injected into the context window incurs both monetary cost and attention dilution. To understand the severity of this problem, we conduct a large-scale empirical study of 55,315 publicly available skills and find systemic inefficiencies: 26.4% lack routing descriptions entirely, over 60% of body content is non-actionable, and reference files can inject tens of thousands of tokens per invocation. Motivated by these findings, we present textscSkillReducer, a two-stage optimization framework. Stage~1 optimizes the routing layer by compressing verbose descriptions and generating missing ones via adversarial delta debugging. Stage~2 restructures skill bodies through taxonomy-driven classification and progressive disclosure, separating actionable core rules from supplementary content loaded on demand, validated by faithfulness checks and a self-correcting feedback loop. Evaluated on 600 skills and the SkillsBench benchmark, textscSkillReducer achieves 48% description compression and 39% body compression while improving functional quality by 2.8%, revealing a less-is-more effect where removing non-essential content reduces distraction in the context window. These benefits transfer across five models from four families with a mean retention of 0.965, and generalize to an independent agent framework.

Key facts. Cited by 29 papers in the map (data through 7 Oct 2026). Placed in Agent skill learning and evolution, one of 16 directions in LLM agents and tool use. Builds on 24 of its 33 references that are in the map. First posted on arXiv 31 Mar 2026.

Builds on

24 of 33 references.

Closest ancestors · 1

Same direction, newest first

  1. 2026SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks13 Feb

Other references · 23

  1. 2020Retrieval-Augmented Generation for Knowledge-Intensive NLP Taskscited by 9,160
  2. 2025Qwen3 Technical Reportcited by 8,195
  3. 2024Qwen2.5 Technical Reportcited by 4,713
  4. 2023Lost in the Middle: How Language Models Use Long Contextscited by 3,383
  5. 2023ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIscited by 1,880
  6. 2023AgentBench: Evaluating LLMs as Agentscited by 1,211
  7. 2024ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Toolscited by 1,096
  8. 2022Large Language Models Are Human-Level Prompt Engineerscited by 1,020
  9. 2023Large Language Models Can Be Easily Distracted by Irrelevant Contextcited by 820
  10. 2023API-Bank: A Comprehensive Benchmark for Tool-Augmented LLMscited by 574
  11. 2023LongLLMLingua: Accelerating and Enhancing LLMs in Long Context Scenarios via Prompt Compressioncited by 443
  12. 2024Understanding the planning of LLM agents: A surveycited by 414
11 more
  1. 2023Learning to Compress Prompts with Gist Tokenscited by 379
  2. 2023RECOMP: Improving Retrieval-Augmented LMs with Compression and Selective Augmentationcited by 253
  3. 2023Compressing Context to Enhance Inference Efficiency of Large Language Modelscited by 185
  4. 2025Context Length Alone Hurts LLM Performance Despite Perfect Retrievalcited by 118
  5. 2026Agent Skills in the Wild: An Empirical Study of Security Vulnerabilities at Scalecited by 94
  6. 2024Prompt Compression for Large Language Models: A Surveycited by 65
  7. 2025Speed at the Cost of Quality: How Cursor AI Increases Short-Term Velocity and Long-Term Complexity in Open-Source Projectscited by 23
  8. 2024Concise and Precise Context Compression for Tool-Using Language Modelscited by 12
  9. 2024ChatDBG: Augmenting Debugging with Large Language Modelscited by 9
  10. 2018Debloating Software through Piece-Wise Compilation and Loadingcited by 6
  11. 2026AgentCompress: Task-Aware Compression for Affordable Large Language Model Agentscited by 1

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 29.

29 papers.

Sort:

By area:

  1. 2026SkillRevise: Improving LLM-Authored Agent Skills via Trace-Conditioned Skill Revisioncited by 21
  2. 2026More Skills, Worse Agents? Skill Shadowing Degrades Performance When Expanding Skill Librariescited by 11
  3. 2026SkillAxe: Sharpening LLM-Authored Agent Skills Through Evaluation-Guided Self-Refinementcited by 11
  4. 2026OpenSkill: Open-World Self-Evolution for LLM Agentscited by 9
  5. 2026SkillMaster: Toward Autonomous Skill Mastery in LLM Agentscited by 8
  6. 2026Managing Procedural Memory in LLM Agents: Control, Adaptation, and Evaluationcited by 8
  7. 2026Agent Skill Evaluation and Evolution: Frameworks and Benchmarkscited by 6
  8. 2026Skill Self-Play: Pushing the Frontier of LLM Capability with Co-Evolving Skillscited by 6
  9. 2026From Anatomy to Smells: An Empirical Study of SKILL.md in Agent Skillscited by 5
  10. 2026SkillZip: Evaluation-Free Skill Compression for Self-Evolving Agents by Discovering Reusable Structurecited by 3
  11. 2026Ace-Skill: Bootstrapping Multimodal Agents with Prioritized and Clustered Evolutioncited by 2
  12. 2026Attacking the Trusted Imagination: Oracle-Level Integrity Attacks on Imagine-then-Act World Modelscited by 2
17 more
  1. 2026SKILL-KD: Contrastive Skill Distillation for LLM Agentscited by 2
  2. 2026From Relevance to Execution Utility: Reward-Aware Dynamic Execution Gating for Skill-Based LLM Agentscited by 2
  3. 2026Optimal Skill Selection for LLM Agents with Provable Bicriteria Guaranteescited by 2
  4. 2026ObjectGraph: From Document Injection to Knowledge Traversal -- A Native File Format for the Agentic Eracited by 1
  5. 2026"Skill Issues'': Data-Centric Optimization of Lakehouse Agentscited by 1
  6. 2026SkillZip: Contract-Preserving Graph Compression for Scalable Agent Skill Librariescited by 1
  7. 2026What Keeps Agent Skills from Being Reusable? Evidence from 138K SKILL.md Filescited by 1
  8. 2026SkillGate: Training In-Policy Skill Selection in Long-Horizon Agentscited by 1
  9. 2026The Router Within: Eliciting Native Skill Routing from a Frozen LLMcited by 1
  10. 2026SkillPager: Query-Adaptive Intra-Skill Navigation via Semantic Node Retrievalcited by 0
  11. 2026Compression, structure, and executor capability: a controlled real-cost decomposition of language-model agent skill optimisationcited by 0
  12. 2026VCE-Skill: Enhancing Skill Self-Evolution with Version-Change Experiencecited by 0
  13. 2026Towards a Systems Foundation for Agentic Skills: Architecture, Lifecycle, and Securitycited by 0
  14. 2026SkillZip Pro: Execution-Aware Dynamic Compression of Progressively Loaded Skills for Self-Evolving Agentscited by 0
  15. 2026Demystifying Agent Skills for Smart Contract Auditing: Design, Effectiveness, Behavioral Impactcited by 0
  16. 2026Safe Skill Retirement for Physical Agentscited by 0
  17. 2026SkillSeek: Revisiting Agent Skill Retrieval at Marketplace Scalecited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026SkillRouter: Skill Routing for LLM Agents at Scale
  2. 2026SkillBloat: Token Amplification Attacks via Skill Injection in LLM Coding Agents
  3. 2026SkillGrad: Optimizing Agent Skills Like Gradient Descent
  4. 2026GraSP: Graph-Structured Skill Compositions for LLM Agents
  5. 2026SkillRet: A Large-Scale Benchmark for Skill Retrieval in LLM Agents
  6. 2026LatentSkill: From In-Context Textual Skills to In-Weight Latent Skills for LLM Agents
  7. 2026SkCC: Portable and Secure Skill Compilation for Cross-Framework LLM Agents
  8. 2026SkillsInjector: Dynamic Skill Context Construction for LLM Agents