{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.18051", "kind": "paper", "url": "https://arxiv.org/abs/2606.18051"}], "authors": ["Xueping Gao"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.18051", "source_url": "https://arxiv.org/abs/2606.18051"}, "categories": ["General AI", "Language & Knowledge", "cs.CL"], "description": {"en": "CompSkillBench evaluates compositional skill routing for LLM agents: given a user query and a library of 2,209 real MCP server skills across 24 categories, systems must decompose the query into sub-tasks, retrieve a skill per sub-task, and produce an executable plan. Scoring uses step-level category recall and decomposition accuracy."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.18051", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.18051", "title": "CompSkillBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Alibaba Group (China)"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.18051", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.18051", "source_url": "https://openalex.org/W7164986448"}, "key": "claire-radar:2606.18051", "modality": null, "name": "CompSkillBench", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "High", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-10-14T09:44:51Z", "data_status": "primary-source-indexed", "display_eligible": "unknown", "origin_record_id": "bm_compskillbench_264db8aa", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "aba4c83ff4974bb44182871ab6d69548416d76a34a0e4677e171fde352c73fde", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.18051"}, "publisher": null, "released": "2026-06-16", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.18051", "source_url": "https://arxiv.org/abs/2606.18051"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-18051", "source": "claire_radar", "source_benchmark_id": "2606.18051", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Language & Knowledge", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": null, "githubScope": null, "githubStars": null, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 1, "hfPaperUrl": "https://huggingface.co/papers/2606.18051", "observedAt": "2026-10-04T03:02:42.754940Z", "signalStatus": {"githubStars": {"state": "not_applicable"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-10-02", "state": "stale"}}, "status": "stale"}, "capabilities": [], "confidence": "High", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "553c75df1ad6e3823e7c6253733a0e05ffa540f0c43794b319282f8b7d7dbac3", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "CompSkillBench evaluates compositional skill routing for LLM agents: given a user query and a library of 2,209 real MCP server skills across 24 categories, systems must decompose the query into sub-tasks, retrieve a skill per sub-task, and produce an executable plan. Scoring uses step-level category recall and decomposition accuracy.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact named benchmark artifact released in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence"], "snippet": "To support evaluation, we introduce CompSkillBench, a benchmark of 300 compositional queries over 2,209 real MCP server skills spanning 24 functional categories, sourced from the public MCP ecosystem."}, "familyId": "bmf_6de15eac705a", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_compskillbench_264db8aa", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2606.18051", "paper": "https://arxiv.org/abs/2606.18051", "pdf": "https://arxiv.org/pdf/2606.18051", "project": null, "report": "https://arxiv.org/abs/2606.18051"}, "metrics": [], "motivation": "LLM agents increasingly rely on external skills -- reusable tool specifications -- but real-world tasks often require composing multiple skills, not just selecting one.", "name": "CompSkillBench", "oneLine": "CompSkillBench evaluates compositional skill routing for LLM agents: given a user query and a library of 2,209 real MCP server skills across 24 categories, systems must decompose the query into sub-tasks, retrieve a skill per sub-task, and produce an executable plan. Scoring uses step-level category recall and decomposition accuracy.", "paperTitle": "Compositional Skill Routing for LLM Agents: Decompose, Retrieve, and Compose", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.18051", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "SkillWeaver Team", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://arxiv.org/abs/2606.18051"}], "ranking": {}, "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-16", "source": {"authors": ["Xueping Gao"], "categories": ["cs.CL"], "id": "2606.18051", "title": "Compositional Skill Routing for LLM Agents: Decompose, Retrieve, and Compose", "type": "arxiv", "url": "https://arxiv.org/abs/2606.18051"}, "sourceUpdatedAt": "2026-06-16T15:27:55Z", "topics": ["cs.CL"], "venueAttempts": [], "whyItMatters": "Real-world agent tasks often require composing multiple tools, but existing benchmarks emphasize single-skill selection. CompSkillBench provides a reusable, ecosystem-grounded benchmark to measure decomposition quality and retrieval in a compositional setting, helping developers identify bottlenecks in agent pipelines—particularly the critical role of task decomposition before retrieval."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
