{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.02540", "kind": "paper", "url": "https://arxiv.org/abs/2606.02540"}, {"id": "gh:osu-nlp-group/skillharm", "kind": "repo", "url": "https://github.com/OSU-NLP-Group/SkillHarm"}], "authors": ["Yuting Ning", "Zhehao Zhang", "Yash Kumar Lal", "Boyu Gou", "Junyi Li", "Weitong Ruan", "Chentao Ye", "Rahul Gupta", "Diyi Yang", "Yu Su", "Huan Sun"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.02540", "source_url": "https://arxiv.org/abs/2606.02540"}, "categories": ["General AI", "Language & Knowledge", "cs.CL"], "description": {"en": "Benchmark of skill-based attacks across the skill-use lifecycle, with 879 attack samples across 71 skills, evaluating Fixed-Payload Poisoning and Self-Mutating Poisoning scenarios across 12 risk types, with attack success rate as primary metric."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.02540", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.02540", "title": "SkillHarm"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.02540", "modality": null, "name": "SkillHarm", "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_skillharm_560fd4c5", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "13c71e8f8e6e03088c616f8517c3358deee5dba2132db52d5a58cec5d1956294", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.02540"}, "publisher": null, "released": "2026-06-01", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.02540", "source_url": "https://arxiv.org/abs/2606.02540"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-02540", "source": "claire_radar", "source_benchmark_id": "2606.02540", "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/OSU-NLP-Group/SkillHarm", "githubScope": "benchmark_repo", "githubStars": 10, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 10, "hfPaperUrl": "https://huggingface.co/papers/2606.02540", "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": "2026-09-30", "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": "d9bf68d429d90f900361bb66d90fcee3ebc365ed25421becd42c88b9fb1e5dc7", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "Benchmark of skill-based attacks across the skill-use lifecycle, with 879 attack samples across 71 skills, evaluating Fixed-Payload Poisoning and Self-Mutating Poisoning scenarios across 12 risk types, with attack success rate as primary metric.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "benchmark release stated in one sentence", "evaluation protocol evidence"], "snippet": "To bridge these gaps, we introduce SkillHarm, a benchmark of skill-based attacks across the skill-use lifecycle, paired with a systematic taxonomy of skill-relevant risks."}, "familyId": "bmf_8422ad323405", "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/OSU-NLP-Group/SkillHarm", "url": "https://github.com/OSU-NLP-Group/SkillHarm"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/OSU-NLP-Group/SkillHarm", "url": "https://huggingface.co/papers/2606.02540"}], "scope": "benchmark_repo", "url": "https://github.com/OSU-NLP-Group/SkillHarm", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_skillharm_560fd4c5", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/OSU-NLP-Group/SkillHarm", "data": null, "hfPaper": "https://huggingface.co/papers/2606.02540", "paper": "https://arxiv.org/abs/2606.02540", "pdf": "https://arxiv.org/pdf/2606.02540", "project": null, "report": "https://arxiv.org/abs/2606.02540"}, "metrics": [], "motivation": "Agent skills occupy a privileged position in the agent workflow, as agents are expected to implicitly follow and execute them, rendering third-party skills a vulnerable attack surface.", "name": "SkillHarm", "oneLine": "Benchmark of skill-based attacks across the skill-use lifecycle, with 879 attack samples across 71 skills, evaluating Fixed-Payload Poisoning and Self-Mutating Poisoning scenarios across 12 risk types, with attack success rate as primary metric.", "paperTitle": "SkillHarm: Lifecycle-Aware Skill-Based Attacks via Automated Construction", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.02540", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "OSU-NLP-Group", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/OSU-NLP-Group/SkillHarm"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-01", "source": {"authors": ["Yuting Ning", "Zhehao Zhang", "Yash Kumar Lal", "Boyu Gou", "Junyi Li", "Weitong Ruan", "Chentao Ye", "Rahul Gupta", "Diyi Yang", "Yu Su", "Huan Sun"], "categories": ["cs.CL"], "id": "2606.02540", "title": "SkillHarm: Lifecycle-Aware Skill-Based Attacks via Automated Construction", "type": "arxiv", "url": "https://arxiv.org/abs/2606.02540"}, "sourceUpdatedAt": "2026-06-01T17:45:39Z", "topics": ["cs.CL"], "venueAttempts": [], "whyItMatters": "Existing studies evaluate poisoned skills within single task executions and use ad-hoc risk lists. This benchmark systematically covers lifecycle-aware attacks and provides a taxonomy and construction pipeline for reproducible evaluation."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
