{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2607.09711", "kind": "paper", "url": "https://arxiv.org/abs/2607.09711"}], "authors": ["Zhiyuan Peng", "Xin Yin", "Chenhao Ying", "Zhe Cui", "Zixiang Ding", "Zhenhua Liu", "Jiang Wu", "Yuan Luo"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2607.09711", "source_url": "https://arxiv.org/abs/2607.09711"}, "categories": ["General AI", "Language & Knowledge", "cs.LG"], "description": {"en": "EvoClawBench evaluates whether an agent runtime can convert evidence from its own runs into reusable skills that improve fresh executions. It covers 100 tasks and 502 sub-problems across coding, data, office, security, operations, and domain-document workflows, comparing direct execution, pre-authored skills, and post-run skill summarization."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2607.09711", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2607.09711", "title": "EvoClawBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Shanghai Jiao Tong University", "Zhejiang University"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2607.09711", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2607.09711", "source_url": "https://openalex.org/W7168292687"}, "key": "claire-radar:2607.09711", "modality": null, "name": "EvoClawBench", "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_evoclawbench_03620568", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "805b72959a32a77aa4e9975f7cf4b43abf57f1b2ca630afbb0a5b654596146f9", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2607.09711"}, "publisher": null, "released": "2026-06-23", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2607.09711", "source_url": "https://arxiv.org/abs/2607.09711"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2607-09711", "source": "claire_radar", "source_benchmark_id": "2607.09711", "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/2607.09711", "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": "2026-10-04", "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": "245f3c84838aebaf1465c81f18340a4346f4ad460b6505d9bda65fc4b1e648ec", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "EvoClawBench evaluates whether an agent runtime can convert evidence from its own runs into reusable skills that improve fresh executions. It covers 100 tasks and 502 sub-problems across coding, data, office, security, operations, and domain-document workflows, comparing direct execution, pre-authored skills, and post-run skill summarization.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "benchmark release stated in one sentence", "evaluation protocol evidence"], "snippet": "We introduce EvoClawBench, a benchmark for this closed-loop skill-learning question on repeated, fixture-backed tasks."}, "familyId": "bmf_3bf0a17e8b5d", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_evoclawbench_03620568", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2607.09711", "paper": "https://arxiv.org/abs/2607.09711", "pdf": "https://arxiv.org/pdf/2607.09711", "project": null, "report": "https://arxiv.org/abs/2607.09711"}, "metrics": [], "motivation": "Existing agent benchmarks primarily test task completion, tool use, or skill utility, but do not isolate whether a runtime can convert evidence from its own runs into reusable skills that improve fresh executions after authoring overhead.", "name": "EvoClawBench", "oneLine": "EvoClawBench evaluates whether an agent runtime can convert evidence from its own runs into reusable skills that improve fresh executions. It covers 100 tasks and 502 sub-problems across coding, data, office, security, operations, and domain-document workflows, comparing direct execution, pre-authored skills, and post-run skill summarization.", "paperTitle": "EvoClawBench: Can Agents Learn Reusable Skills from Their Own Runs?", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2607.09711", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {}, "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-23", "source": {"authors": ["Zhiyuan Peng", "Xin Yin", "Chenhao Ying", "Zhe Cui", "Zixiang Ding", "Zhenhua Liu", "Jiang Wu", "Yuan Luo"], "categories": ["cs.LG", "cs.SE"], "id": "2607.09711", "title": "EvoClawBench: Can Agents Learn Reusable Skills from Their Own Runs?", "type": "arxiv", "url": "https://arxiv.org/abs/2607.09711"}, "sourceUpdatedAt": "2026-06-23T19:14:29Z", "topics": ["cs.LG"], "venueAttempts": [], "whyItMatters": "The evaluation gap is assessing closed-loop skill learning in agents, where benefits are selective and cost-sensitive rather than automatic. The benchmark provides a decision value for runtime developers and users considering skill-authoring loops."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
