{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2608.19901", "kind": "paper", "url": "https://arxiv.org/abs/2608.19901"}, {"kind": "website", "url": "https://protectskills.github.io/MaliciousSkillBench/"}], "authors": ["Yue Wang", "Yi Liu", "Gelei Deng", "Ying Zhang", "Yuekang Li", "Zhenyu Chen", "Leo Zhang"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2608.19901", "source_url": "https://arxiv.org/abs/2608.19901"}, "categories": ["Agents", "General AI", "Vision & 3D"], "description": {"en": "Benchmark for detecting malicious agent skills, comprising 9,740 skills (7,505 malicious, 2,235 benign) and evaluating detectors under random, structural-disjoint, and source-disjoint splits."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2608.19901", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2608.19901", "title": "MaliciousSkillBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Nanjing University", "Griffith University", "Nanyang Technological University", "Wake Forest University", "UNSW Sydney"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2608.19901", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2608.19901", "source_url": "https://openalex.org/W7203938462"}, "key": "claire-radar:2608.19901", "modality": null, "name": "MaliciousSkillBench", "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": "true", "origin_record_id": "bm_maliciousskillbench_08ed3dc5", "origin_source": "arxiv", "recognition_confidence": "0.95", "record_sha256": "b8e4b261b3d0b3253d50ec9c7885b36da294494ad371c2478ff58ff118b4b6f5", "relation": "introduces", "review_model": "deepseek-v4-pro", "review_state": "ai-reviewed", "reviewed_at": "2026-08-26T06:15:55.708965Z", "source_url": "https://arxiv.org/abs/2608.19901"}, "publisher": null, "released": "2026-08-20", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2608.19901", "source_url": "https://arxiv.org/abs/2608.19901"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2608-19901", "source": "claire_radar", "source_benchmark_id": "2608.19901", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Vision & 3D", "attention": {"asOf": "2026-08-26", "attemptedAt": "2026-08-25T23:36:26.005677Z", "githubRepo": null, "githubScope": null, "githubStars": null, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 0, "hfPaperUrl": "https://huggingface.co/papers/2608.19901", "observedAt": "2026-08-25T23:36:26.005677Z", "signalStatus": {"githubStars": {"state": "not_applicable"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"observedAt": "2026-08-25T23:36:26.005677Z", "state": "fresh"}}, "status": "fresh"}, "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "The benchmark addresses security risks in agent skills, a timely topic, and includes comprehensive evaluation results likely to draw interest from the AI safety community.", "score": 55}, "capabilities": [], "confidence": "High", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-26T06:15:55.708965Z", "inputHash": "26519b25e53beac4cec813dd1535878febdf57535c076da579e4cd40a5d10762", "model": "deepseek-v4-pro"}, "curation": {"canonicalNameEvidence": "MaliciousSkillBench: A Comprehensive Benchmark for Malicious Agent Skill Detection", "canonicalNameSource": "paper_title", "decisionReason": "The paper explicitly introduces the benchmark with a stable scoring contract and the official project page signals public availability of resources.", "model": "deepseek-v4-pro", "reviewedAt": "2026-08-26T06:15:55.708965Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "Benchmark for detecting malicious agent skills, comprising 9,740 skills (7,505 malicious, 2,235 benign) and evaluating detectors under random, structural-disjoint, and source-disjoint splits.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "benchmark release stated in one sentence", "public artifact URL"], "snippet": "We present MaliciousSkillBench, a comprehensive benchmark for malicious Agent Skill detection."}, "familyId": "bmf_0285cbc8ba0d", "firstSeenAt": "2026-08-21", "heat": null, "id": "bm_maliciousskillbench_08ed3dc5", "indexedAt": "2026-08-21T04:30:09.569171Z", "industrySectors": [], "links": {"code": null, "data": null, "paper": "https://arxiv.org/abs/2608.19901", "pdf": "https://arxiv.org/pdf/2608.19901", "project": "https://protectskills.github.io/MaliciousSkillBench/", "report": "https://arxiv.org/abs/2608.19901"}, "metrics": [], "motivation": "Agent Skills extend LLM agents with reusable instruction packages that may also include scripts, resources, and service configuration.", "name": "MaliciousSkillBench", "oneLine": "Benchmark for detecting malicious agent skills, comprising 9,740 skills (7,505 malicious, 2,235 benign) and evaluating detectors under random, structural-disjoint, and source-disjoint splits.", "paperTitle": "MaliciousSkillBench: A Comprehensive Benchmark for Malicious Agent Skill Detection", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2608.19901", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-21T04:30:09.569171Z"}, "publications": [], "publishers": [{"name": "ProtectSkills", "organizationType": "benchmark-organization", "role": "benchmark-publisher", "sourceUrl": "https://protectskills.github.io/MaliciousSkillBench/"}], "ranking": {"30d": {"confidence": "Low", "coverage": 0.3, "growth": {"components": {"githubStars": {"percentile": null, "value": null}, "hfDatasetDownloads": {"percentile": null, "value": null}, "hfPaperUpvotes": {"percentile": null, "value": null}}, "confidence": "Low", "coverage": 0.0, "elapsedDays": null, "historyDate": null, "rank": null, "score": null, "windowDays": 30}, "level": {"components": {"githubStars": {"percentile": null, "value": null}, "hfDatasetDownloads": {"percentile": null, "value": null}, "hfPaperUpvotes": {"percentile": 0.1712962962962963, "value": 0}}, "confidence": "Low", "coverage": 0.3, "rank": null, "score": 40}, "method": "attention-ranking-v11", "rank": null, "score": 40}, "90d": {"confidence": "Low", "coverage": 0.15, "growth": {"components": {"githubStars": {"percentile": null, "value": null}, "hfDatasetDownloads": {"percentile": null, "value": null}, "hfPaperUpvotes": {"percentile": null, "value": null}}, "confidence": "Low", "coverage": 0.0, "elapsedDays": null, "historyDate": null, "rank": null, "score": null, "windowDays": 90}, "level": {"components": {"githubStars": {"percentile": null, "value": null}, "hfDatasetDownloads": {"percentile": null, "value": null}, "hfPaperUpvotes": {"percentile": 0.20813953488372092, "value": 0}}, "confidence": "Low", "coverage": 0.15, "rank": null, "score": 46}, "method": "attention-ranking-v11", "rank": null, "score": 46}}, "readiness": "Paper only", "recognitionConfidence": 0.95, "relation": "introduces", "releasedAt": "2026-08-20", "source": {"authors": ["Yue Wang", "Yi Liu", "Gelei Deng", "Ying Zhang", "Yuekang Li", "Zhenyu Chen", "Leo Zhang"], "categories": ["cs.CR", "cs.AI"], "id": "2608.19901", "title": "MaliciousSkillBench: A Comprehensive Benchmark for Malicious Agent Skill Detection", "type": "arxiv", "url": "https://arxiv.org/abs/2608.19901"}, "sourceUpdatedAt": "2026-08-20T11:13:00Z", "topics": ["Agents"], "venueAttempts": [], "whyItMatters": "Addresses the need for robust malicious skill detection by providing a consolidated cross-source dataset and evaluation protocols that measure both detection and false-positive rates."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
