{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.15899", "kind": "paper", "url": "https://arxiv.org/abs/2606.15899"}], "authors": ["Ismail Hossain", "Sai Puppala", "Md Jahangir Alam", "Tanzim Ahad", "Sajedul Talukder"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.15899", "source_url": "https://arxiv.org/abs/2606.15899"}, "categories": ["Agents", "General AI", "Language & Knowledge"], "description": {"en": "SkillVetBench evaluates security risks in open-source LLM agent skills through multidimensional LLM-based judging and a weighted agentic-risk score."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.15899", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.15899", "title": "SkillVetBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["The University of Texas at El Paso"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.15899", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.15899", "source_url": "https://openalex.org/W7164889658"}, "key": "claire-radar:2606.15899", "modality": null, "name": "SkillVetBench", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"admission_policy_version": "2026-09-09.1", "confidence": "Medium", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-09-25T04:40:22Z", "data_status": "primary-source-indexed", "display_eligible": "true", "origin_record_id": "bm_skillvetbench_abedf80f", "origin_source": "arxiv", "recognition_confidence": "0.65", "record_sha256": "112e743dad9eacd5b7ab0a4144a847dda80292401a372d6652e1307677d5ed2c", "relation": "introduces", "review_model": "codex-cli-default", "review_state": "ai-reviewed", "reviewed_at": "2026-09-24T17:43:24.279638Z", "source_url": "https://arxiv.org/abs/2606.15899"}, "publisher": null, "released": "2026-06-14", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.15899", "source_url": "https://arxiv.org/abs/2606.15899"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-15899", "source": "claire_radar", "source_benchmark_id": "2606.15899", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Language & Knowledge", "attentionForecast": {"confidence": "Low", "horizon": "7d", "reason": "Agent-skill security and public leaderboard comparisons may attract focused near-term attention, though the evidence does not establish broader uptake.", "score": 58}, "capabilities": [], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T17:43:24.279638Z", "inputHash": "af120f2796fb2c03679a59fa3383dc1b6cd2c00b3107e6c5d91c6e6342b812dc", "model": "codex-cli-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "SkillVetBench: LLM-as-Judge for Multi-Dimensional Security Risk Evaluation", "canonicalNameSource": "paper_title", "decisionReason": "Publish because the source names a repeatable evaluation, defines a multidimensional scoring metric, and identifies a live public leaderboard as a reuse and comparison path.", "evaluationEvidence": {"reuse": "a live public leaderboard on Hugging Face", "scoring": "SARS (Skill Agentic Risk Score), a five-dimensional agentic-risk metric with a principled weighted formula", "task": "uses an LLM-as-Judge to vet agent skills"}, "model": "codex-cli-default", "reviewedAt": "2026-09-24T17:43:24.279638Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "SkillVetBench evaluates security risks in open-source LLM agent skills through multidimensional LLM-based judging and a weighted agentic-risk score.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "score_submission", "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "evaluation protocol evidence"], "snippet": "We present SKILLVETBENCH, a live public leaderboard on Hugging Face that uses an LLM-as-Judge to vet agent skills."}, "familyId": "bmf_13b32737c884", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_skillvetbench_abedf80f", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": null, "paper": "https://arxiv.org/abs/2606.15899", "pdf": "https://arxiv.org/pdf/2606.15899", "project": null, "report": "https://arxiv.org/abs/2606.15899"}, "metrics": [], "motivation": "Open-source LLM agent ecosystems are growing rapidly, yet the security of community-contributed skills - modular tool definitions that extend agent capabilities - remains largely unvetted.", "name": "SkillVetBench", "oneLine": "SkillVetBench evaluates security risks in open-source LLM agent skills through multidimensional LLM-based judging and a weighted agentic-risk score.", "paperTitle": "SkillVetBench: LLM-as-Judge for Multi-Dimensional Security Risk Evaluation in Open-Source LLM Agent Skills", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.15899", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "readiness": "Paper only", "recognitionConfidence": 0.65, "relation": "introduces", "releasedAt": "2026-06-14", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["Ismail Hossain", "Sai Puppala", "Md Jahangir Alam", "Tanzim Ahad", "and Sajedul Talukder"], "categories": ["cs.CR", "cs.AI", "cs.HC", "cs.LG", "cs.MA"], "id": "2606.15899", "title": "SkillVetBench: LLM-as-Judge for Multi-Dimensional Security Risk Evaluation in Open-Source LLM Agent Skills", "type": "arxiv", "url": "https://arxiv.org/abs/2606.15899"}, "sourceUpdatedAt": "2026-06-14T16:30:33Z", "topics": ["Agents"], "venueAttempts": [], "whyItMatters": "It targets instruction-layer and multi-agent security risks that code-layer scanners may not detect."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
