{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2607.08768", "kind": "paper", "url": "https://arxiv.org/abs/2607.08768"}, {"id": "gh:hku-mmlab/uniclawbench", "kind": "repo", "url": "https://github.com/HKU-MMLab/UniClawBench"}, {"kind": "website", "url": "https://uniclawbench.github.io"}], "authors": ["Zhekai Chen", "Chengqi Duan", "Kaiyue Sun", "Bohao Li", "Yuqing Wang", "Manyuan Zhang", "Xihui Liu"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2607.08768", "source_url": "https://arxiv.org/abs/2607.08768"}, "categories": ["General AI", "Language & Knowledge", "cs.CL"], "description": {"en": "A capability-driven benchmark for proactive agents in real-world tasks, with 400 bilingual tasks across five capabilities. It evaluates agents in Docker containers using step-by-step checkpoints and a closed-loop strategy with executor, supervisor, and user agents."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2607.08768", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2607.08768", "title": "UniClawBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2607.08768", "modality": null, "name": "UniClawBench", "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_uniclawbench_4d04906d", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "d1f86b96d25a9d8ba2bb01d654965607ff437583f898b83cd9828236c601a1fd", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2607.08768"}, "publisher": null, "released": "2026-07-09", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2607.08768", "source_url": "https://arxiv.org/abs/2607.08768"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2607-08768", "source": "claire_radar", "source_benchmark_id": "2607.08768", "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": "https://github.com/HKU-MMLab/UniClawBench", "githubScope": "benchmark_repo", "githubStars": 39, "hfDailySubmittedAt": "2026-07-10T00:00:00.000Z", "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 34, "hfPaperUrl": "https://huggingface.co/papers/2607.08768", "observedAt": "2026-10-04T15:31:48.030916Z", "signalStatus": {"githubStars": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}}, "status": "fresh"}, "capabilities": [], "confidence": "High", "construction": "Interactive Environment", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "2f779a2c4ca85aa66b2e5e837c8b79ccc20b727c7fcd6b6b7d04b6ff494ace17", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "A capability-driven benchmark for proactive agents in real-world tasks, with 400 bilingual tasks across five capabilities. It evaluates agents in Docker containers using step-by-step checkpoints and a closed-loop strategy with executor, supervisor, and user agents.", "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", "public artifact URL"], "snippet": "To address these limitations, we introduce UniClawBench, the first capability-driven benchmark designed to evaluate proactive agents in dynamic, real-world settings."}, "familyId": "bmf_36cef9ecb058", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [{"evidence": [{"basis": "existing-code-link", "linkedUrl": "https://github.com/HKU-MMLab/UniClawBench", "url": "https://github.com/HKU-MMLab/UniClawBench"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/HKU-MMLab/UniClawBench", "url": "https://huggingface.co/papers/2607.08768"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/HKU-MMLab/UniClawBench.", "url": "https://arxiv.org/abs/2607.08768"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/HKU-MMLab/UniClawBench", "url": "https://arxiv.org/abs/2607.08768"}], "scope": "benchmark_repo", "url": "https://github.com/HKU-MMLab/UniClawBench", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_uniclawbench_4d04906d", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/HKU-MMLab/UniClawBench", "data": null, "hfPaper": "https://huggingface.co/papers/2607.08768", "paper": "https://arxiv.org/abs/2607.08768", "pdf": "https://arxiv.org/pdf/2607.08768", "project": "https://uniclawbench.github.io", "report": "https://arxiv.org/abs/2607.08768"}, "metrics": [], "motivation": "The rapid development of large language models and multimodal large language models has accelerated the emergence of proactive agents capable of operating everyday tools and assisting users in real-world environments.", "name": "UniClawBench", "oneLine": "A capability-driven benchmark for proactive agents in real-world tasks, with 400 bilingual tasks across five capabilities. It evaluates agents in Docker containers using step-by-step checkpoints and a closed-loop strategy with executor, supervisor, and user agents.", "paperTitle": "UniClawBench: A Universal Benchmark for Proactive Agents on Real-World Tasks", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2607.08768", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "HKU-MMLab", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/HKU-MMLab/UniClawBench"}], "ranking": {"90d": {"confidence": "Medium", "coverage": 0.7, "growth": {"components": {"githubStars": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfDatasetDownloads": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfPaperUpvotes": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}}, "confidence": "Low", "coverage": 0.0, "elapsedDays": null, "historyDate": null, "rank": null, "score": null, "windowDays": 90}, "level": {"components": {"githubStars": {"normalization": "log-max", "normalized": 0.49239909538041343, "value": 39}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": 0.6282108373075276, "value": 34}}, "confidence": "Medium", "coverage": 0.7, "rank": 53, "score": 52}, "method": "attention-ranking-v13", "rank": 53, "score": 52}}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-07-09", "source": {"authors": ["Zhekai Chen", "Chengqi Duan", "Kaiyue Sun", "Bohao Li", "Yuqing Wang", "Manyuan Zhang", "Xihui Liu"], "categories": ["cs.CL"], "id": "2607.08768", "title": "UniClawBench: A Universal Benchmark for Proactive Agents on Real-World Tasks", "type": "arxiv", "url": "https://arxiv.org/abs/2607.08768"}, "sourceUpdatedAt": "2026-07-09T17:59:32Z", "topics": ["cs.CL"], "venueAttempts": [], "whyItMatters": "Existing agent benchmarks rely on sandboxed environments and single-turn paradigms, which do not reflect real-world complexity. This benchmark provides a dynamic, capability-based evaluation that helps compare models and agent frameworks, aiding in identifying failure root causes."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
