{"work":{"id":"b477d12a-2ca6-4894-9f90-3fb479635e98","openalex_id":null,"doi":null,"arxiv_id":"2602.12670","raw_key":null,"title":"SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks","authors":null,"authors_text":"Xiangyi Li, Wenbo Chen, Yimin Liu, Shenghan Zheng, Xiaokun Chen, Yifeng He","year":2026,"venue":"cs.AI","abstract":"Agent Skills are structured packages of procedural knowledge that augment LLM agents at inference time. Despite rapid adoption, there is no standard way to measure whether they actually help. We present SkillsBench, a benchmark of 86 tasks across 11 domains paired with curated Skills and deterministic verifiers. Each task is evaluated under three conditions: no Skills, curated Skills, and self-generated Skills. We test 7 agent-model configurations over 7,308 trajectories. Curated Skills raise average pass rate by 16.2 percentage points(pp), but effects vary widely by domain (+4.5pp for Software Engineering to +51.9pp for Healthcare) and 16 of 84 tasks show negative deltas. Self-generated Skills provide no benefit on average, showing that models cannot reliably author the procedural knowledge they benefit from consuming. Focused Skills with 2--3 modules outperform comprehensive documentation, and smaller models with Skills can match larger models without them.","external_url":"https://arxiv.org/abs/2602.12670","cited_by_count":null,"metadata_source":"pith","metadata_fetched_at":"2026-05-25T06:50:28.630073+00:00","pith_arxiv_id":"2602.12670","created_at":"2026-05-09T06:10:36.654330+00:00","updated_at":"2026-06-05T21:23:00.469572+00:00","title_quality_ok":true,"display_title":"SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks","render_title":"SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks"},"hub":{"state":{"work_id":"b477d12a-2ca6-4894-9f90-3fb479635e98","tier":"hub","tier_reason":"10+ Pith inbound or 1,000+ external citations","pith_inbound_count":53,"external_cited_by_count":null,"distinct_field_count":10,"first_pith_cited_at":"2026-02-24T13:11:38+00:00","last_pith_cited_at":"2026-05-22T17:59:50+00:00","author_build_status":"not_needed","summary_status":"needed","contexts_status":"needed","graph_status":"needed","ask_index_status":"not_needed","reader_status":"not_needed","recognition_status":"not_needed","updated_at":"2026-06-12T01:59:10.926770+00:00","tier_text":"hub"},"tier":"hub","role_counts":[{"context_role":"background","n":13},{"context_role":"dataset","n":2},{"context_role":"baseline","n":1},{"context_role":"other","n":1}],"polarity_counts":[{"context_polarity":"background","n":13},{"context_polarity":"unclear","n":2},{"context_polarity":"baseline","n":1},{"context_polarity":"use_dataset","n":1}],"runs":{"context_extract":{"job_type":"context_extract","status":"succeeded","result":{"enqueued_papers":25},"error":null,"updated_at":"2026-05-14T18:09:41.975181+00:00"},"graph_features":{"job_type":"graph_features","status":"succeeded","result":{"co_cited":[{"title":"Voyager: An Open-Ended Embodied Agent with Large Language Models","work_id":"ffe0d207-86cf-4742-a100-e988ac8b9676","shared_citers":13},{"title":"Agent Skills for Large Language Models: Architecture, Acquisition, Security, and the Path Forward","work_id":"a64d3985-6826-492c-8a2e-f0965d805bfd","shared_citers":10},{"title":"MemSkill: Learning and Evolving Memory Skills for Self-Evolving Agents","work_id":"64baa7f3-849b-485e-b081-2074d82f1364","shared_citers":10},{"title":"arXiv preprint arXiv:2603.02176 , year=","work_id":"da16cf2d-c19f-4d7a-8fa3-78fa0c9117a8","shared_citers":9},{"title":"SkillRL: Evolving Agents via Recursive Skill-Augmented Reinforcement Learning","work_id":"536a9d0a-baed-4eb9-9a51-f2b585c3388b","shared_citers":9},{"title":"SoK: Agentic skills–beyond tool use in LLM agents","work_id":"db16e56a-bf5c-4b9b-ab0e-2c76449f4dcb","shared_citers":9},{"title":"arXiv preprint arXiv:2603.04448 , year=","work_id":"47ef6124-7b5c-4acd-94d6-cbd6ab854c9c","shared_citers":8},{"title":"Fatemi, Xiaolong Jin, Zora Zhiruo Wang, Apurva Gandhi, Yueqi Song, Yu Gu, Jayanth Srinivasa, Gaowen Liu, Graham Neubig, and Yu Su","work_id":"72b1aff0-91c3-4500-abc9-22129ca67748","shared_citers":8},{"title":"arXiv preprint arXiv:2603.01145 , year=","work_id":"f5786621-deeb-44fb-9f12-c1d70df052ea","shared_citers":7},{"title":"Evoskill: Automated skill discovery for multi-agent systems.arXiv preprint arXiv:2603.02766","work_id":"94c2c7b2-0c94-4b32-90cc-bc154d192850","shared_citers":7},{"title":"GPT-4 Technical Report","work_id":"b928e041-6991-4c08-8c81-0359e4097c7b","shared_citers":7},{"title":"Spatialagent: An autonomous ai agent for spatial biology.bioRxiv, pp","work_id":"48e4c885-a8df-43e2-8b95-4a17ec085323","shared_citers":7},{"title":"arXiv preprint arXiv:2603.12056 (2026)","work_id":"6bec6eef-86ec-49a5-a730-a01e19656d75","shared_citers":6},{"title":"Memento-skills: Let agents design agents","work_id":"53800d8e-8b59-418c-b6cb-3cd56186c42e","shared_citers":6},{"title":"ReAct: Synergizing Reasoning and Acting in Language Models","work_id":"407a2351-25f1-497d-b611-f77d0292a8e6","shared_citers":6},{"title":"SkillRouter: Retrieve-and-rerank skill selection for LLM agents at scale","work_id":"07f70041-6a66-424d-ba53-819c03e42308","shared_citers":6},{"title":"Swe-skills-bench: Do agent skills actually help in real-world software engineering?","work_id":"960dedfb-e8af-4c41-bba8-b565f908ed71","shared_citers":6},{"title":"CoEvoSkills: Self-Evolving Agent Skills via Co-Evolutionary Verification","work_id":"509f2c7c-7ee4-4b35-aa68-a3db839ad987","shared_citers":5},{"title":"DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models","work_id":"c5006563-f3ec-438a-9e35-b7b484f34828","shared_citers":5},{"title":"Trace2Skill: Distill Trajectory-Local Lessons into Transferable Agent Skills","work_id":"c6f8f524-e13a-48b6-99b8-7dd3357d9665","shared_citers":5},{"title":"AgentBench: Evaluating LLMs as Agents","work_id":"a37549b4-4c94-412d-acc4-4efeb08509be","shared_citers":4},{"title":"Agent skills in the wild: An empirical study of security vulnerabilities at scale","work_id":"0780d782-7977-4d63-a1f2-3ae587a49c98","shared_citers":4},{"title":"arXiv preprint arXiv:2511.20857 , year=","work_id":"eafad83d-679f-4020-8413-9451109c58ba","shared_citers":4},{"title":"arXiv preprint arXiv:2602.08004 , year=","work_id":"36e2cb78-60dc-4d3b-8857-15b16900172d","shared_citers":4}],"time_series":[{"n":35,"year":2026}],"dependency_candidates":[]},"error":null,"updated_at":"2026-05-14T18:09:42.006618+00:00"},"identity_refresh":{"job_type":"identity_refresh","status":"succeeded","result":{"items":[{"title":"Qwen3 Technical Report","outcome":"unchanged","work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e","resolver":"local_arxiv","confidence":0.98,"old_work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e"}],"counts":{"fixed":0,"merged":0,"unchanged":1,"quarantined":0,"needs_external_resolution":0},"errors":[],"attempted":1},"error":null,"updated_at":"2026-05-14T18:09:57.847414+00:00"},"summary_claims":{"job_type":"summary_claims","status":"succeeded","result":{"title":"SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks","claims":[{"claim_text":"Agent Skills are structured packages of procedural knowledge that augment LLM agents at inference time. Despite rapid adoption, there is no standard way to measure whether they actually help. We present SkillsBench, a benchmark of 86 tasks across 11 domains paired with curated Skills and deterministic verifiers. Each task is evaluated under three conditions: no Skills, curated Skills, and self-generated Skills. We test 7 agent-model configurations over 7,308 trajectories. Curated Skills raise average pass rate by 16.2 percentage points(pp), but effects vary widely by domain (+4.5pp for Softwar","claim_type":"abstract","evidence_strength":"source_metadata"}],"why_cited":"Pith tracks SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks because it crossed a citation-hub threshold.","role_counts":[]},"error":null,"updated_at":"2026-05-14T18:09:19.762396+00:00"}},"summary":{"title":"SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks","claims":[{"claim_text":"Agent Skills are structured packages of procedural knowledge that augment LLM agents at inference time. Despite rapid adoption, there is no standard way to measure whether they actually help. We present SkillsBench, a benchmark of 86 tasks across 11 domains paired with curated Skills and deterministic verifiers. Each task is evaluated under three conditions: no Skills, curated Skills, and self-generated Skills. We test 7 agent-model configurations over 7,308 trajectories. Curated Skills raise average pass rate by 16.2 percentage points(pp), but effects vary widely by domain (+4.5pp for Softwar","claim_type":"abstract","evidence_strength":"source_metadata"}],"why_cited":"Pith tracks SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks because it crossed a citation-hub threshold.","role_counts":[]},"graph":{"co_cited":[{"title":"Voyager: An Open-Ended Embodied Agent with Large Language Models","work_id":"ffe0d207-86cf-4742-a100-e988ac8b9676","shared_citers":13},{"title":"Agent Skills for Large Language Models: Architecture, Acquisition, Security, and the Path Forward","work_id":"a64d3985-6826-492c-8a2e-f0965d805bfd","shared_citers":10},{"title":"MemSkill: Learning and Evolving Memory Skills for Self-Evolving Agents","work_id":"64baa7f3-849b-485e-b081-2074d82f1364","shared_citers":10},{"title":"arXiv preprint arXiv:2603.02176 , year=","work_id":"da16cf2d-c19f-4d7a-8fa3-78fa0c9117a8","shared_citers":9},{"title":"SkillRL: Evolving Agents via Recursive Skill-Augmented Reinforcement Learning","work_id":"536a9d0a-baed-4eb9-9a51-f2b585c3388b","shared_citers":9},{"title":"SoK: Agentic skills–beyond tool use in LLM agents","work_id":"db16e56a-bf5c-4b9b-ab0e-2c76449f4dcb","shared_citers":9},{"title":"arXiv preprint arXiv:2603.04448 , year=","work_id":"47ef6124-7b5c-4acd-94d6-cbd6ab854c9c","shared_citers":8},{"title":"Fatemi, Xiaolong Jin, Zora Zhiruo Wang, Apurva Gandhi, Yueqi Song, Yu Gu, Jayanth Srinivasa, Gaowen Liu, Graham Neubig, and Yu Su","work_id":"72b1aff0-91c3-4500-abc9-22129ca67748","shared_citers":8},{"title":"arXiv preprint arXiv:2603.01145 , year=","work_id":"f5786621-deeb-44fb-9f12-c1d70df052ea","shared_citers":7},{"title":"Evoskill: Automated skill discovery for multi-agent systems.arXiv preprint arXiv:2603.02766","work_id":"94c2c7b2-0c94-4b32-90cc-bc154d192850","shared_citers":7},{"title":"GPT-4 Technical Report","work_id":"b928e041-6991-4c08-8c81-0359e4097c7b","shared_citers":7},{"title":"Spatialagent: An autonomous ai agent for spatial biology.bioRxiv, pp","work_id":"48e4c885-a8df-43e2-8b95-4a17ec085323","shared_citers":7},{"title":"arXiv preprint arXiv:2603.12056 (2026)","work_id":"6bec6eef-86ec-49a5-a730-a01e19656d75","shared_citers":6},{"title":"Memento-skills: Let agents design agents","work_id":"53800d8e-8b59-418c-b6cb-3cd56186c42e","shared_citers":6},{"title":"ReAct: Synergizing Reasoning and Acting in Language Models","work_id":"407a2351-25f1-497d-b611-f77d0292a8e6","shared_citers":6},{"title":"SkillRouter: Retrieve-and-rerank skill selection for LLM agents at scale","work_id":"07f70041-6a66-424d-ba53-819c03e42308","shared_citers":6},{"title":"Swe-skills-bench: Do agent skills actually help in real-world software engineering?","work_id":"960dedfb-e8af-4c41-bba8-b565f908ed71","shared_citers":6},{"title":"CoEvoSkills: Self-Evolving Agent Skills via Co-Evolutionary Verification","work_id":"509f2c7c-7ee4-4b35-aa68-a3db839ad987","shared_citers":5},{"title":"DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models","work_id":"c5006563-f3ec-438a-9e35-b7b484f34828","shared_citers":5},{"title":"Trace2Skill: Distill Trajectory-Local Lessons into Transferable Agent Skills","work_id":"c6f8f524-e13a-48b6-99b8-7dd3357d9665","shared_citers":5},{"title":"AgentBench: Evaluating LLMs as Agents","work_id":"a37549b4-4c94-412d-acc4-4efeb08509be","shared_citers":4},{"title":"Agent skills in the wild: An empirical study of security vulnerabilities at scale","work_id":"0780d782-7977-4d63-a1f2-3ae587a49c98","shared_citers":4},{"title":"arXiv preprint arXiv:2511.20857 , year=","work_id":"eafad83d-679f-4020-8413-9451109c58ba","shared_citers":4},{"title":"arXiv preprint arXiv:2602.08004 , year=","work_id":"36e2cb78-60dc-4d3b-8857-15b16900172d","shared_citers":4}],"time_series":[{"n":35,"year":2026}],"dependency_candidates":[]},"authors":[]}}