{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2608.02670", "kind": "paper", "url": "https://arxiv.org/abs/2608.02670"}], "authors": ["Dotan Davidovich", "Yair Amar", "Hai Rozencwajg", "Or Hiltch"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2608.02670", "source_url": "https://arxiv.org/abs/2608.02670"}, "categories": ["General AI", "Language & Knowledge", "cs.CR"], "description": {"en": "Boundary-Bench is an open-source plugin that adds configurable security policy levels to Terminal-Bench, enabling evaluation of coding agents under constraints like scoped credentials, restricted egress, read-only filesystems, and non-root execution."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2608.02670", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2608.02670", "title": "Boundary-Bench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2608.02670", "modality": null, "name": "Boundary-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_boundary-bench_5415f6d9", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "749c8c6edde56d4722974bd54ca9d45a7e26cf8ae125373ffa15bf25a2839e93", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2608.02670"}, "publisher": null, "released": "2026-08-02", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2608.02670", "source_url": "https://arxiv.org/abs/2608.02670"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2608-02670", "source": "claire_radar", "source_benchmark_id": "2608.02670", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Mixed", "applicationDomains": ["General AI"], "area": "Language & Knowledge", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": null, "githubScope": null, "githubStars": null, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 3, "hfPaperUrl": "https://huggingface.co/papers/2608.02670", "observedAt": "2026-10-04T15:31:48.030916Z", "signalStatus": {"githubStars": {"state": "not_applicable"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}}, "status": "fresh"}, "capabilities": [], "confidence": "High", "construction": "Aggregate Existing", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "9a2d68a81eda969254befd1930f2ad4b0a24d9426358d5cb66bb11f832e4ce48", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "Boundary-Bench is an open-source plugin that adds configurable security policy levels to Terminal-Bench, enabling evaluation of coding agents under constraints like scoped credentials, restricted egress, read-only filesystems, and non-root execution.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact named benchmark artifact released in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence"], "snippet": "We release Boundary-Bench, an open-source hardening plugin enabling policy-constrained evaluation of coding agents on Terminal-Bench and compatible benchmarks."}, "familyId": "bmf_29b2f52f6f51", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_boundary-bench_5415f6d9", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2608.02670", "paper": "https://arxiv.org/abs/2608.02670", "pdf": "https://arxiv.org/pdf/2608.02670", "project": null, "report": "https://arxiv.org/abs/2608.02670"}, "metrics": [], "motivation": "Coding agents increasingly run inside organizations whose security controls (scoped credentials, restricted egress, read-only filesystems, non-root execution) constrain them like any other software.", "name": "Boundary-Bench", "oneLine": "Boundary-Bench is an open-source plugin that adds configurable security policy levels to Terminal-Bench, enabling evaluation of coding agents under constraints like scoped credentials, restricted egress, read-only filesystems, and non-root execution.", "paperTitle": "Permission Denied: Policy-Graded Evaluation of Coding Agents in Hardened Environments", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2608.02670", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "Boundary-Bench authors", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://arxiv.org/abs/2608.02670"}], "ranking": {"90d": {"confidence": "Low", "coverage": 0.15, "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": null, "value": null}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": 0.24495074076910422, "value": 3}}, "confidence": "Low", "coverage": 0.15, "rank": null, "score": 46}, "method": "attention-ranking-v13", "rank": null, "score": 46}}, "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-08-02", "source": {"authors": ["Dotan Davidovich", "Yair Amar", "Hai Rozencwajg", "Or Hiltch"], "categories": ["cs.CR", "cs.AI"], "id": "2608.02670", "title": "Permission Denied: Policy-Graded Evaluation of Coding Agents in Hardened Environments", "type": "arxiv", "url": "https://arxiv.org/abs/2608.02670"}, "sourceUpdatedAt": "2026-08-02T14:26:12Z", "topics": ["cs.CR"], "venueAttempts": [], "whyItMatters": "Existing coding agent benchmarks assume permissive sandboxes, leaving a gap in understanding performance under real-world security policies. This benchmark provides a standardized way to measure success and efficiency trade-offs across policy levels, informing model selection for deployment in hardened environments."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
