{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.15242", "kind": "paper", "url": "https://arxiv.org/abs/2606.15242"}, {"id": "gh:saint-viperx/scr_bench", "kind": "repo", "url": "https://github.com/saint-viperx/SCR_Bench"}], "authors": ["Yi Xie", "Jiawei Du", "Yu Cheng", "Jiuan Zhou", "Zhaoxia Yin"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.15242", "source_url": "https://arxiv.org/abs/2606.15242"}, "categories": ["Agents", "General AI", "Language & Knowledge"], "description": {"en": "SCR-Bench is a benchmark for evaluating security risks in composed LLM agent skill workflows. It includes three sub-benchmarks (SCR-CapFlow, SCR-TrustLift, SCR-AuthBlur) that measure attack success rates, trust transfer, and authorization confusion in sandboxed environments."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.15242", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.15242", "title": "SCR-Bench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["East China Normal University", "A*STAR Graduate Academy", "Shanghai Innovation Institute"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.15242", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.15242", "source_url": "https://openalex.org/W7164930602"}, "key": "claire-radar:2606.15242", "modality": null, "name": "SCR-Bench", "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_scr-bench_984f0fe8", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "1f21b9fdc292bd0bad1a316642238609c9b108811bdf32d7c571d4c7614ec175", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.15242"}, "publisher": null, "released": "2026-06-13", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.15242", "source_url": "https://arxiv.org/abs/2606.15242"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-15242", "source": "claire_radar", "source_benchmark_id": "2606.15242", "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/saint-viperx/SCR_Bench", "githubScope": "benchmark_repo", "githubStars": 16, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 1, "hfPaperUrl": "https://huggingface.co/papers/2606.15242", "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": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-10-01", "state": "stale"}}, "status": "fresh"}, "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": "aa29672772df59586e700ba9db9bc6f2001ea20ee085d0daea2c6c162a3a8f47", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "SCR-Bench is a benchmark for evaluating security risks in composed LLM agent skill workflows. It includes three sub-benchmarks (SCR-CapFlow, SCR-TrustLift, SCR-AuthBlur) that measure attack success rates, trust transfer, and authorization confusion in sandboxed environments.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact named benchmark artifact released in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "We introduce SCR-Bench to evaluate this risk in controlled, sandboxed skill environments."}, "familyId": "bmf_ae50c3042f6a", "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/saint-viperx/SCR_Bench", "url": "https://github.com/saint-viperx/SCR_Bench"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/saint-viperx/SCR_Bench", "url": "https://huggingface.co/papers/2606.15242"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/saint-viperx/SCR_Bench.", "url": "https://arxiv.org/abs/2606.15242"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/saint-viperx/SCR_Bench", "url": "https://arxiv.org/abs/2606.15242"}], "scope": "benchmark_repo", "url": "https://github.com/saint-viperx/SCR_Bench", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_scr-bench_984f0fe8", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/saint-viperx/SCR_Bench", "data": null, "hfPaper": "https://huggingface.co/papers/2606.15242", "paper": "https://arxiv.org/abs/2606.15242", "pdf": "https://arxiv.org/pdf/2606.15242", "project": null, "report": "https://arxiv.org/abs/2606.15242"}, "metrics": [], "motivation": "Skills are becoming the capability layer through which LLM agents turn plans into actions, but their use introduces security risks such as data leakage, unauthorized operations, and tool misuse.", "name": "SCR-Bench", "oneLine": "SCR-Bench is a benchmark for evaluating security risks in composed LLM agent skill workflows. It includes three sub-benchmarks (SCR-CapFlow, SCR-TrustLift, SCR-AuthBlur) that measure attack success rates, trust transfer, and authorization confusion in sandboxed environments.", "paperTitle": "Benign in Isolation, Harmful in Composition: Security Risks in Agent Skill Ecosystems", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.15242", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "SCR-Bench Team", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/saint-viperx/SCR_Bench"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-13", "source": {"authors": ["Yi Xie", "Jiawei Du", "Yu Cheng", "Jiuan Zhou", "Zhaoxia Yin"], "categories": ["cs.CR", "cs.AI"], "id": "2606.15242", "title": "Benign in Isolation, Harmful in Composition: Security Risks in Agent Skill Ecosystems", "type": "arxiv", "url": "https://arxiv.org/abs/2606.15242"}, "sourceUpdatedAt": "2026-06-13T10:55:01Z", "topics": ["Agents"], "venueAttempts": [], "whyItMatters": "SCR-Bench addresses the gap in evaluating agent skill security at the path level, where skills benign in isolation become harmful in composition. It provides a controlled, sandboxed environment for assessing composition-induced risks."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
