{"aliases": [], "authors": ["Chang Ma", "Junlei Zhang", "Zhihao Zhu", "Cheng Yang", "Yujiu Yang", "Yaohui Jin", "Zhenzhong Lan", "Lingpeng Kong", "Junxian He"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "22c5a60f9ed7a8488ec31fbf24d392e9674b0cae79ab1b269fd9d57202025801", "submitted_at": "2024-01-24T01:51:00Z", "updated_at": "2024-12-23T20:12:48Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2401.13178v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:31:24Z", "role": "primary", "sha256": "3aac6e9748769efff3c475ede0da192776daa03431d231b5f67e2aa50f8a3be9", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2308.03688%2C2308.15272%2C2309.08172%2C2309.11436%2C2310.04716%2C2310.04869%2C2310.05126%2C2310.06770%2C2310.08740%2C2310.10634%2C2310.11441%2C2310.16042%2C2310.18331%2C2311.01767%2C2311.07562%2C2311.09278%2C2311.12983%2C2311.18751%2C2312.03003%2C2312.06677%2C2312.08914%2C2312.11671%2C2312.13108%2C2312.13771%2C2312.15820%2C2312.16062%2C2401.01614%2C2401.04124%2C2401.07781%2C2401.10935%2C2401.13178%2C2401.13649%2C2401.13919%2C2401.14268%2C2401.16158%2C2402.01030%2C2402.04476%2C2402.04615%2C2402.05930%2C2402.06596%2C2402.07456%2C2402.07939%2C2402.07945%2C2402.08567%2C2402.11941%2C2402.14073%2C2402.15057%2C2402.16965%2C2402.17553%2C2403.02713&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:31:24Z", "role": "primary", "sha256": "3aac6e9748769efff3c475ede0da192776daa03431d231b5f67e2aa50f8a3be9", "supports": ["name", "kind", "summary", "topics", "notes"], "url": "https://export.arxiv.org/api/query?id_list=2308.03688%2C2308.15272%2C2309.08172%2C2309.11436%2C2310.04716%2C2310.04869%2C2310.05126%2C2310.06770%2C2310.08740%2C2310.10634%2C2310.11441%2C2310.16042%2C2310.18331%2C2311.01767%2C2311.07562%2C2311.09278%2C2311.12983%2C2311.18751%2C2312.03003%2C2312.06677%2C2312.08914%2C2312.11671%2C2312.13108%2C2312.13771%2C2312.15820%2C2312.16062%2C2401.01614%2C2401.04124%2C2401.07781%2C2401.10935%2C2401.13178%2C2401.13649%2C2401.13919%2C2401.14268%2C2401.16158%2C2402.01030%2C2402.04476%2C2402.04615%2C2402.05930%2C2402.06596%2C2402.07456%2C2402.07939%2C2402.07945%2C2402.08567%2C2402.11941%2C2402.14073%2C2402.15057%2C2402.16965%2C2402.17553%2C2403.02713&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T09:15:30Z", "role": "primary", "sha256": "bbfef23bd53d2febb258a91d5a706f630cd597909bdbea56f16ac97dbe4777c0", "supports": ["summary", "environments", "scope", "notes"], "url": "https://arxiv.org/html/2401.13178v2"}], "id": "arxiv:2401.13178", "identifiers": {"arxiv": "2401.13178"}, "kind": "paper", "last_verified_at": "2026-10-16T06:31:24Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2401.13178"}], "name": "AgentBoard: An Analytical Evaluation Board of Multi-turn LLM Agents", "notes": "Reviewed against the retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced. The full paper documents search/click WebShop actions and 11 browser actions in WebArena, while the overall benchmark also covers non-web domains.", "published_at": "2024-01-24", "relations": [], "review_status": "curated", "scope": "adjacent", "summary": "Introduces AgentBoard, a multiturn agent evaluation framework with intermediate progress metrics and analysis tools. Its broader task suite includes WebShop and WebArena interactions with stepwise progress measurements.", "topics": ["evaluation", "observability", "web-navigation"], "updated_at": "2024-12-23", "url": "https://arxiv.org/abs/2401.13178"}
{"aliases": [], "authors": ["Eric Li", "Jim Waldo"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "b9a728edf0b834adb72abfc50e13d5b2b52f532f444b3e8af34effd3d86beb1b", "submitted_at": "2024-06-01T00:32:26Z", "updated_at": "2024-06-01T00:32:26Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2406.01623v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:58Z", "role": "primary", "sha256": "c87e97eb397da04806d53145cf9c13be739fc657bb54dca9318c8c4ed2a36249", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=800&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:58Z", "role": "primary", "sha256": "c87e97eb397da04806d53145cf9c13be739fc657bb54dca9318c8c4ed2a36249", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=800&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2406.01623", "identifiers": {"arxiv": "2406.01623"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:58Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2406.01623"}], "name": "WebSuite: Systematically Evaluating Why Web Agents Fail", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2024-06-01", "relations": [], "review_status": "curated", "scope": "core", "summary": "WebSuite organizes browser operations into an action taxonomy and tests both isolated controls and complete workflows. Its diagnostic tasks attribute failures to specific actions, enabling comparison of text-based and multimodal web agents.", "topics": ["evaluation", "observability", "web-navigation"], "updated_at": "2024-06-01", "url": "https://arxiv.org/abs/2406.01623"}
{"aliases": [], "authors": ["Reyna Abhyankar", "Qi Qi", "Yiying Zhang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "df3f654ef1364aeb6cdaeaea1c6334da949a798b2f8256c9ddc4e6a88a588e19", "submitted_at": "2025-06-19T05:26:40Z", "updated_at": "2026-05-18T17:55:59Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2506.16042v2"}}, "environments": ["desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2506.16042", "identifiers": {"arxiv": "2506.16042"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2506.16042"}], "name": "OSWorld-Human: Benchmarking the Efficiency of Computer-Use Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-06-19", "relations": [], "review_status": "curated", "scope": "core", "summary": "OSWorld-Human adds human-determined trajectories to OSWorld to evaluate the efficiency of desktop computer-use agents. The study compares 16 agents and analyzes model-call latency, changes in step duration during task execution, and excess action steps.", "topics": ["evaluation", "efficiency", "observability"], "updated_at": "2026-05-18", "url": "https://arxiv.org/abs/2506.16042"}
{"aliases": [], "authors": ["Fengxian Ji", "Jingpu Yang", "Zirui Song", "Yuanxi Wang", "Zhexuan Cui", "Yuke Li", "Qian Jiang", "Miao Fang", "Xiuying Chen"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "f8c91a1f85602760c706e56426c0d2d870bbc101ca500aa3c7c391938c3f1676", "submitted_at": "2025-08-12T15:12:42Z", "updated_at": "2025-08-12T15:12:42Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2508.09241v1"}}, "environments": ["desktop", "web", "cross-platform"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2508.09241", "identifiers": {"arxiv": "2508.09241"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2508.09241"}], "name": "FineState-Bench: A Comprehensive Benchmark for Fine-Grained State Control in GUI Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-08-12", "relations": [], "review_status": "curated", "scope": "core", "summary": "FineState-Bench evaluates fine-grained GUI perception and control across desktop, web, and mobile interfaces. Its Visual Diagnostic Assistant separates perception and localization contributions in controlled interaction experiments.", "topics": ["evaluation", "grounding", "observability", "screen-understanding"], "updated_at": "2025-08-12", "url": "https://arxiv.org/abs/2508.09241"}
{"aliases": [], "authors": ["Yutong Bian", "Xianhao Lin", "Yupeng Xie", "Tianyang Liu", "Mingchen Zhuge", "Siyuan Lu", "Haoming Tang", "Jinlin Wang", "Jiayi Zhang", "Jiaqi Chen", "Xiangru Tang", "Yongxin Ni", "Sirui Hong", "Chenglin Wu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "c5034c06e63607a1279668964e91c14ed762a95b6e3953337ef8b8b63c4fd57d", "submitted_at": "2025-08-17T07:31:11Z", "updated_at": "2025-08-17T07:31:11Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2508.14104v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:32:52Z", "role": "primary", "sha256": "0b730323b2ce85c39aa7d534665901a480ff06540c73f1f80832fcc0735b8789", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2310.08588%2C2402.09205%2C2406.05720%2C2407.21075%2C2410.07166%2C2503.03743%2C2508.04037%2C2508.10833%2C2508.11360%2C2508.13024%2C2508.13186%2C2508.14104%2C2509.01619%2C2509.01720%2C2509.02322%2C2509.04198%2C2509.07764%2C2509.21823%2C2509.22256%2C2509.23263%2C2509.23866%2C2509.25047%2C2509.26539%2C2510.00413%2C2510.00536%2C2510.01524%2C2510.02204%2C2510.02250%2C2510.02418%2C2510.03204%2C2510.03612%2C2510.04039%2C2510.04373%2C2510.04607%2C2510.06223%2C2510.06607%2C2510.09872%2C2510.10666%2C2510.11035%2C2510.11221%2C2510.11892%2C2510.13543%2C2510.15455%2C2510.15863%2C2510.16252%2C2510.18113%2C2510.18314%2C2510.19205%2C2510.22732%2C2511.06101&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:32:52Z", "role": "primary", "sha256": "0b730323b2ce85c39aa7d534665901a480ff06540c73f1f80832fcc0735b8789", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2310.08588%2C2402.09205%2C2406.05720%2C2407.21075%2C2410.07166%2C2503.03743%2C2508.04037%2C2508.10833%2C2508.11360%2C2508.13024%2C2508.13186%2C2508.14104%2C2509.01619%2C2509.01720%2C2509.02322%2C2509.04198%2C2509.07764%2C2509.21823%2C2509.22256%2C2509.23263%2C2509.23866%2C2509.25047%2C2509.26539%2C2510.00413%2C2510.00536%2C2510.01524%2C2510.02204%2C2510.02250%2C2510.02418%2C2510.03204%2C2510.03612%2C2510.04039%2C2510.04373%2C2510.04607%2C2510.06223%2C2510.06607%2C2510.09872%2C2510.10666%2C2510.11035%2C2510.11221%2C2510.11892%2C2510.13543%2C2510.15455%2C2510.15863%2C2510.16252%2C2510.18113%2C2510.18314%2C2510.19205%2C2510.22732%2C2511.06101&max_results=50"}], "id": "arxiv:2508.14104", "identifiers": {"arxiv": "2508.14104"}, "kind": "paper", "last_verified_at": "2026-10-16T06:32:52Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2508.14104"}], "name": "You Don't Know Until You Click:Automated GUI Testing for Production-Ready Software Evaluation", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-08-17", "relations": [], "review_status": "curated", "scope": "core", "summary": "RealDevWorld evaluates generated software repositories with 194 development tasks and AppEvalPilot, an agent judge that operates the resulting GUIs. The evaluation examines functional behavior, visual fidelity, and runtime responses to produce task-specific diagnostic feedback.", "topics": ["evaluation", "automation", "observability"], "updated_at": "2025-08-17", "url": "https://arxiv.org/abs/2508.14104"}
{"aliases": [], "authors": ["Haitao Hu", "Peng Chen", "Yanpeng Zhao", "Yuqi Chen"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "ae582aceb31075c460a9e4a6dce9978574c3e0ada731b26c13738a4a4fafcc32", "submitted_at": "2025-09-09T13:59:00Z", "updated_at": "2025-09-09T13:59:00Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2509.07764v1"}}, "environments": ["desktop", "linux", "web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "topics", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T11:25:42Z", "role": "primary", "sha256": "7f303629ef6d9c7df128c6d12d472abd9bf238e0204515f18839249adef317b9", "supports": ["summary", "environments", "scope", "notes"], "url": "https://arxiv.org/html/2509.07764v1"}], "id": "arxiv:2509.07764", "identifiers": {"arxiv": "2509.07764"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2509.07764"}], "name": "AgentSentinel: An End-to-End and Real-Time Security Defense Framework for Computer-Use Agents", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced. The official full paper confirms GUI/desktop interaction alongside Bash and file tools. Its abstract and body enumerate attack categories differently; no category count is promoted into this annotation.", "published_at": "2025-09-09", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces AgentSentinel, which intercepts sensitive computer-agent operations and audits them using task context together with system traces. BadComputerUse evaluates this enforcement approach across attack scenarios involving tool and computer control.", "topics": ["safety-security", "observability", "evaluation"], "updated_at": "2025-09-09", "url": "https://arxiv.org/abs/2509.07764"}
{"aliases": [], "authors": ["Daniel Röder", "Akhil Juneja", "Roland Roller", "Sven Schmeier"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "ac93e8cfca06ff86fad06408051abaa6b6510a44d7ad9f2d34c478b11391a32f", "submitted_at": "2025-09-17T19:34:49Z", "updated_at": "2025-09-17T19:34:49Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2509.14382v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2509.14382", "identifiers": {"arxiv": "2509.14382"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2509.14382"}], "name": "Detecting Pipeline Failures through Fine-Grained Analysis of Web Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-09-17", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces a modular framework for identifying errors at intermediate stages of a web-agent pipeline. A SeeAct and Mind2Web case study examines weaknesses that are obscured by final task-success metrics.", "topics": ["evaluation", "observability", "web-navigation"], "updated_at": "2025-09-17", "url": "https://arxiv.org/abs/2509.14382"}
{"aliases": [], "authors": ["Jiazheng Sun", "Ruimeng Yang", "Xu Han", "Jiayang Niu", "Mingxuan Li", "Te Yang", "Yongyong Lu", "Xin Peng"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "571eee6476c9e75ffae65cf98ec589e98fd3b953313cc6d2191bb48bd27303fe", "submitted_at": "2025-09-25T04:21:31Z", "updated_at": "2025-12-01T08:01:05Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2509.20729v2"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2509.20729", "identifiers": {"arxiv": "2509.20729"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2509.20729"}], "name": "Robust, Observable, and Evolvable Agentic Systems Engineering: A Principled Framework Validated via the Fairy GUI Agent", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2025-09-25", "relations": [], "review_status": "curated", "scope": "core", "summary": "Presents an agent-engineering framework combining runtime goal clarification, an observable component architecture and evolving execution memory. The Fairy mobile GUI agent evaluates these mechanisms on ambiguous tasks and through controlled and user studies.", "topics": ["observability", "human-in-the-loop", "memory", "robustness"], "updated_at": "2025-12-01", "url": "https://arxiv.org/abs/2509.20729"}
{"aliases": [], "authors": ["Yaoyao Qian", "Yuanli Wang", "Jinda Zhang", "Yun Zong", "Meixu Chen", "Hanhan Zhou", "Jindan Huang", "Yifan Zeng", "Xinyu Hu", "Chan Hee Song", "Danqing Zhang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "d404ff922ef38e57481c01a5e3108bd38457d88f83d7adf6d3c3c7b5ddf124e2", "submitted_at": "2025-10-22T03:29:25Z", "updated_at": "2025-10-22T03:29:25Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2510.19205v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2510.19205", "identifiers": {"arxiv": "2510.19205"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:47Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2510.19205"}], "name": "WebGraphEval: Multi-Turn Trajectory Evaluation for Web Agents using Graph Representation", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2025-10-22", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces WebGraphEval, which merges multiple agents' trajectories into a weighted graph of canonical actions. Structural analyses examine recurring behavior, inefficient paths and important decision points beyond a final success score.", "topics": ["evaluation", "observability", "interaction-history"], "updated_at": "2025-10-22", "url": "https://arxiv.org/abs/2510.19205"}
{"aliases": [], "authors": ["Leyang Yang", "Ziwei Wang", "Xiaoxuan Tang", "Sheng Zhou", "Dajun Chen", "Wei Jiang", "Yong Li"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "dc7f5f0fe1ca373272cda4bbeca3a465e1a7eaec829e8a59bbecbe6049dc7b4a", "submitted_at": "2025-11-12T09:49:31Z", "updated_at": "2025-11-12T09:49:31Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2511.09157v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2511.09157", "identifiers": {"arxiv": "2511.09157"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:47Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2511.09157"}], "name": "ProBench: Benchmarking GUI Agents with Accurate Process Information", "notes": "The annotation was reviewed against the complete retained primary abstract. Experiments and released artifacts have not been independently reproduced.", "published_at": "2025-11-12", "relations": [], "review_status": "curated", "scope": "core", "summary": "ProBench extends mobile GUI evaluation with tasks whose correctness depends on the interaction process as well as the final state. Its ProcessProvider supplies process information for assessing agent behavior and analyzing failure modes.", "topics": ["evaluation", "mobile-navigation", "observability"], "updated_at": "2025-11-12", "url": "https://arxiv.org/abs/2511.09157"}
{"aliases": [], "authors": ["Kevin Qinghong Lin", "Siyuan Hu", "Linjie Li", "Zhengyuan Yang", "Lijuan Wang", "Philip Torr", "Mike Zheng Shou"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "7d63831bc0ff00979615415ce94abc140b646a4375cb33b950c4a7ccf00bf166", "submitted_at": "2025-11-19T16:00:02Z", "updated_at": "2025-11-19T16:00:02Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2511.15567v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2511.15567", "identifiers": {"arxiv": "2511.15567"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:47Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2511.15567"}], "name": "Computer-Use Agents as Judges for Generative User Interface", "notes": "The annotation was reviewed against the complete retained primary abstract. Experiments and released artifacts have not been independently reproduced.", "published_at": "2025-11-19", "relations": [], "review_status": "curated", "scope": "core", "summary": "AUI-Gym evaluates generated websites by having computer-use agents attempt tasks whose executability is checked programmatically. A feedback dashboard condenses navigation histories into visual summaries that a coding agent can use to revise an interface.", "topics": ["evaluation", "ui-design", "observability"], "updated_at": "2025-11-19", "url": "https://arxiv.org/abs/2511.15567"}
{"aliases": [], "authors": ["Youngmin Im", "Byeongung Jo", "Jaeyoung Wi", "Seungwoo Baek", "Tae Hoon Min", "Joo Hyung Lee", "Sangeun Oh", "Insik Shin", "Sunjae Lee"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "5982901598e43dcadf63b4a950b2fff68e85be71d0f483648b535d82d7745bd7", "submitted_at": "2025-12-14T10:41:39Z", "updated_at": "2026-05-31T23:48:10Z", "version": 4, "versioned_url": "https://arxiv.org/abs/2512.12634v4"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2512.12634", "identifiers": {"arxiv": "2512.12634"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:47Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2512.12634"}], "name": "MobiBench: Multi-Branch, Modular Benchmark for Mobile GUI Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-12-14", "relations": [], "review_status": "curated", "scope": "core", "summary": "MobiBench evaluates mobile GUI agents offline while allowing multiple valid action paths. It decomposes agent pipelines into modules to study component contributions, configurations, and performance bottlenecks.", "topics": ["evaluation", "mobile-navigation", "observability"], "updated_at": "2026-05-31", "url": "https://arxiv.org/abs/2512.12634"}
{"aliases": [], "authors": ["Marta Sumyk", "Oleksandr Kosovan"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "048dc7332372585872f5fd3da47a10bd999c7520b2999e45ea6f578b7c856400", "submitted_at": "2026-03-11T09:28:41Z", "updated_at": "2026-03-12T07:30:09Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2603.10577v2"}}, "environments": ["desktop", "macos", "windows", "linux"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2603.10577", "identifiers": {"arxiv": "2603.10577"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:43Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2603.10577"}], "name": "CUAAudit: Meta-Evaluation of Vision-Language Models as Auditors of Autonomous Computer-Use Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-03-11", "relations": [], "review_status": "curated", "scope": "core", "summary": "CUAAudit examines vision-language models that judge computer-use task completion from an instruction and final environment state. Across macOS, Windows and Linux benchmarks, it measures accuracy, confidence calibration and disagreement between auditors.", "topics": ["evaluation", "observability", "robustness"], "updated_at": "2026-03-12", "url": "https://arxiv.org/abs/2603.10577"}
{"aliases": [], "authors": ["Alexandre De Masi"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "5944045ee4562641ee02612969d1c2d038e019fa02cc8d48c671c1d5397557ad", "submitted_at": "2026-03-11T11:24:31Z", "updated_at": "2026-03-11T11:24:31Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2603.10664v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-17T12:31:16Z", "role": "primary", "sha256": "9289bc047b59e93f75703b72f5e92d9277e115238427519c343df550220e351b", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2602.19870%2C2602.22769%2C2602.23166%2C2602.23320%2C2603.00198%2C2603.00842%2C2603.00991%2C2603.01050%2C2603.01145%2C2603.02236%2C2603.02795%2C2603.03278%2C2603.04549%2C2603.06713%2C2603.07106%2C2603.07670%2C2603.07853%2C2603.08706%2C2603.09716%2C2603.10002%2C2603.10664%2C2603.11139%2C2603.11320%2C2603.12483%2C2603.13026%2C2603.13247%2C2603.15714%2C2603.16065%2C2603.16289%2C2603.16839%2C2603.17216%2C2603.17378%2C2603.19313%2C2603.19685%2C2603.21489%2C2603.22527%2C2603.23638%2C2603.24157%2C2603.24639%2C2603.27646%2C2603.28069%2C2603.28407%2C2603.29418%2C2603.29847%2C2604.00528%2C2604.00886%2C2604.01194%2C2604.01438%2C2604.01560%2C2604.02985&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T12:31:16Z", "role": "primary", "sha256": "9289bc047b59e93f75703b72f5e92d9277e115238427519c343df550220e351b", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?id_list=2602.19870%2C2602.22769%2C2602.23166%2C2602.23320%2C2603.00198%2C2603.00842%2C2603.00991%2C2603.01050%2C2603.01145%2C2603.02236%2C2603.02795%2C2603.03278%2C2603.04549%2C2603.06713%2C2603.07106%2C2603.07670%2C2603.07853%2C2603.08706%2C2603.09716%2C2603.10002%2C2603.10664%2C2603.11139%2C2603.11320%2C2603.12483%2C2603.13026%2C2603.13247%2C2603.15714%2C2603.16065%2C2603.16289%2C2603.16839%2C2603.17216%2C2603.17378%2C2603.19313%2C2603.19685%2C2603.21489%2C2603.22527%2C2603.23638%2C2603.24157%2C2603.24639%2C2603.27646%2C2603.28069%2C2603.28407%2C2603.29418%2C2603.29847%2C2604.00528%2C2604.00886%2C2604.01194%2C2604.01438%2C2604.01560%2C2604.02985&max_results=50"}], "id": "arxiv:2603.10664", "identifiers": {"arxiv": "2603.10664"}, "kind": "paper", "last_verified_at": "2026-10-17T12:31:16Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2603.10664"}], "name": "Terminal Is All You Need: Design Properties for Human-AI Agent Collaboration", "notes": "Reviewed against the complete retained official abstract. Experimental findings and release availability have not been independently reproduced or verified. Adjacent conceptual HCI work about agent-facing interfaces; the abstract does not report GUI task experiments or a new evaluated CUA.", "published_at": "2026-03-11", "relations": [], "review_status": "curated", "scope": "adjacent", "summary": "This position paper derives interface design principles from terminal-based human-agent collaboration: representational compatibility, transparent actions and low participation barriers. It argues that graphical and other agent-facing modalities should provide these properties explicitly.", "topics": ["human-in-the-loop", "ui-design", "observability", "tool-integration"], "updated_at": "2026-03-11", "url": "https://arxiv.org/abs/2603.10664"}
{"aliases": [], "authors": ["Haomin Zhuang", "Hanwen Xing", "Xiangliang Zhang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "7b1788e6fbb4c14a6d993ce329bf7bf05e3a1e6ffbfc7c14a9fa9d36fbf7ef16", "submitted_at": "2026-04-15T18:22:50Z", "updated_at": "2026-04-15T18:22:50Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2604.16520v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-17T12:31:19Z", "role": "primary", "sha256": "05ec5d668ac75d3d7c5b71ea7d6e76a3fbf5ebeb742e0843db2329e5e41aa208", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2604.03598%2C2604.04759%2C2604.05547%2C2604.06170%2C2604.07340%2C2604.10547%2C2604.10938%2C2604.11978%2C2604.12282%2C2604.13018%2C2604.13346%2C2604.13954%2C2604.15579%2C2604.15671%2C2604.15715%2C2604.16520%2C2604.18240%2C2604.18934%2C2604.19572%2C2604.19667%2C2604.19742%2C2604.20841%2C2604.20987%2C2604.22436%2C2604.23781%2C2604.24479%2C2604.26148%2C2604.27621%2C2604.27996%2C2604.28093%2C2604.28139%2C2604.28181%2C2605.00425%2C2605.01462%2C2605.02240%2C2605.03042%2C2605.03242%2C2605.04811%2C2605.05191%2C2605.05811%2C2605.06110%2C2605.07073%2C2605.07358%2C2605.08374%2C2605.08985%2C2605.10912%2C2605.10966%2C2605.11363%2C2605.11732%2C2605.11817&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T12:31:19Z", "role": "primary", "sha256": "05ec5d668ac75d3d7c5b71ea7d6e76a3fbf5ebeb742e0843db2329e5e41aa208", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?id_list=2604.03598%2C2604.04759%2C2604.05547%2C2604.06170%2C2604.07340%2C2604.10547%2C2604.10938%2C2604.11978%2C2604.12282%2C2604.13018%2C2604.13346%2C2604.13954%2C2604.15579%2C2604.15671%2C2604.15715%2C2604.16520%2C2604.18240%2C2604.18934%2C2604.19572%2C2604.19667%2C2604.19742%2C2604.20841%2C2604.20987%2C2604.22436%2C2604.23781%2C2604.24479%2C2604.26148%2C2604.27621%2C2604.27996%2C2604.28093%2C2604.28139%2C2604.28181%2C2605.00425%2C2605.01462%2C2605.02240%2C2605.03042%2C2605.03242%2C2605.04811%2C2605.05191%2C2605.05811%2C2605.06110%2C2605.07073%2C2605.07358%2C2605.08374%2C2605.08985%2C2605.10912%2C2605.10966%2C2605.11363%2C2605.11732%2C2605.11817&max_results=50"}], "id": "arxiv:2604.16520", "identifiers": {"arxiv": "2604.16520"}, "kind": "paper", "last_verified_at": "2026-10-17T12:31:19Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2604.16520"}], "name": "AgentClick: A Skill-Based Human-in-the-Loop Review Layer for Terminal AI Agents", "notes": "Reviewed against the complete retained official abstract. Experimental findings and release availability have not been independently reproduced or verified. Adjacent supervision interface: the browser UI is for human review of terminal agents; the abstract does not claim that those agents operate graphical interfaces.", "published_at": "2026-04-15", "relations": [], "review_status": "curated", "scope": "adjacent", "summary": "AgentClick gives terminal-based agents a browser review interface for plans, drafts, execution traces and editable memory. A local server and skill plugin let users inspect proposed actions and provide structured feedback.", "topics": ["human-in-the-loop", "observability", "workflow", "memory", "tool-integration"], "updated_at": "2026-04-15", "url": "https://arxiv.org/abs/2604.16520"}
{"aliases": [], "authors": ["Jinbiao Wei", "Qianran Ma", "Yilun Zhao", "Xiao Zhou", "Kangqi Ni", "Guo Gan", "Arman Cohan"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "0a93d880793018d1b47a39e0487e3be76550fd2b49de9535b2913e9f08159074", "submitted_at": "2026-05-19T12:40:29Z", "updated_at": "2026-05-19T12:40:29Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2605.19769v1"}}, "environments": ["desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2605.19769", "identifiers": {"arxiv": "2605.19769"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:36Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2605.19769"}], "name": "OpenComputer: Verifiable Software Worlds for Computer-Use Agents", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2026-05-19", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces OpenComputer, which builds desktop tasks around application-state verifiers and refines those verifiers using execution feedback. An evaluation harness records trajectories and assigns auditable partial credit across professional software workflows.", "topics": ["evaluation", "data-generation", "observability"], "updated_at": "2026-05-19", "url": "https://arxiv.org/abs/2605.19769"}
{"aliases": [], "authors": ["Bo Zhang", "Borui Zhang", "Chenghao Jiang", "Minglei Shi", "Xiaofeng Wang", "Zheng Zhu", "Jie Zhou", "Jiwen Lu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "33c343e9bc02976fc3e32057cdcdebc3abeae5ef049b877443e81917a8abc550", "submitted_at": "2026-05-28T17:59:31Z", "updated_at": "2026-05-28T17:59:31Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2606.07594v1"}}, "environments": ["desktop", "macos"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-17T12:31:28Z", "role": "primary", "sha256": "7eaa9a5430d3a45049e612a117da345c1daafc54277a35a7cf56b42fbb1e8618", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2606.05646%2C2606.05670%2C2606.06556%2C2606.06891%2C2606.07367%2C2606.07594%2C2606.07682%2C2606.07723%2C2606.07867%2C2606.10394%2C2606.10525%2C2606.11290%2C2606.12203%2C2606.12736%2C2606.13040%2C2606.13368%2C2606.13578%2C2606.13669%2C2606.13802%2C2606.13898%2C2606.14397%2C2606.15057%2C2606.15300%2C2606.15441%2C2606.17162%2C2606.18425%2C2606.19336%2C2606.19409%2C2606.19704%2C2606.20950%2C2606.21033%2C2606.21565%2C2606.23521%2C2606.23654%2C2606.24775%2C2606.25084%2C2606.26080%2C2606.27406%2C2606.27634%2C2606.28430%2C2606.28480%2C2606.29955%2C2606.30616%2C2607.00692%2C2607.01084%2C2607.02032%2C2607.02448%2C2607.02490%2C2607.02577%2C2607.02911&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T12:31:28Z", "role": "primary", "sha256": "7eaa9a5430d3a45049e612a117da345c1daafc54277a35a7cf56b42fbb1e8618", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?id_list=2606.05646%2C2606.05670%2C2606.06556%2C2606.06891%2C2606.07367%2C2606.07594%2C2606.07682%2C2606.07723%2C2606.07867%2C2606.10394%2C2606.10525%2C2606.11290%2C2606.12203%2C2606.12736%2C2606.13040%2C2606.13368%2C2606.13578%2C2606.13669%2C2606.13802%2C2606.13898%2C2606.14397%2C2606.15057%2C2606.15300%2C2606.15441%2C2606.17162%2C2606.18425%2C2606.19336%2C2606.19409%2C2606.19704%2C2606.20950%2C2606.21033%2C2606.21565%2C2606.23521%2C2606.23654%2C2606.24775%2C2606.25084%2C2606.26080%2C2606.27406%2C2606.27634%2C2606.28430%2C2606.28480%2C2606.29955%2C2606.30616%2C2607.00692%2C2607.01084%2C2607.02032%2C2607.02448%2C2607.02490%2C2607.02577%2C2607.02911&max_results=50"}], "id": "arxiv:2606.07594", "identifiers": {"arxiv": "2606.07594"}, "kind": "paper", "last_verified_at": "2026-10-17T12:31:28Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2606.07594"}], "name": "Syll: Open-Source Personal Automation with Cross-Surface Execution", "notes": "Reviewed against the complete retained official abstract. Experimental findings and release availability have not been independently reproduced or verified.", "published_at": "2026-05-28", "relations": [], "review_status": "curated", "scope": "core", "summary": "Syll combines tool calls, shell execution and visual GUI control in a personal-automation runtime. It turns user demonstrations into reusable skills and exposes execution logs, keyframes and approval checkpoints for inspection.", "topics": ["automation", "human-in-the-loop", "workflow", "memory", "recording", "observability"], "updated_at": "2026-05-28", "url": "https://arxiv.org/abs/2606.07594"}
{"aliases": [], "authors": ["Yujin Zhang", "Daye Nam"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "c52695e860206d02735839853416965ec3d42fac134ebfa0e8f2f5f5ea043914", "submitted_at": "2026-06-17T04:13:20Z", "updated_at": "2026-06-17T04:13:20Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2606.18671v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2606.18671", "identifiers": {"arxiv": "2606.18671"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:36Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2606.18671"}], "name": "HANSEL: Extracting Breadcrumbs from Web Agent Trajectories for Interactive Verification", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-06-17", "relations": [], "review_status": "curated", "scope": "core", "summary": "HANSEL extracts evidence pages and snippets from browser-agent traces and preserves relevant page state in interactive verification views. It flags untraceable answers and evaluates evidence selection and human verification effort separately.", "topics": ["human-in-the-loop", "observability", "recording", "evaluation"], "updated_at": "2026-06-17", "url": "https://arxiv.org/abs/2606.18671"}
{"aliases": [], "authors": ["Dayeon Kang", "Hyejun Jeong", "Jade Sheffey", "Pubali Datta", "Amir Houmansadr"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "8b2c6080f0de11c06fe73c30195c7c8529e07de283dc45fa47709072846d385a", "submitted_at": "2026-06-18T20:07:39Z", "updated_at": "2026-06-18T20:07:39Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2606.20910v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2606.20910", "identifiers": {"arxiv": "2606.20910"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:36Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2606.20910"}], "name": "Whose Agent Are You? Multi-Layer Fingerprinting and Attribution of Autonomous Web Agents", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-06-18", "relations": [], "review_status": "curated", "scope": "core", "summary": "Builds a logging framework that combines network/protocol fingerprints with browser interaction behavior to attribute web-agent traffic. A decision-tree classifier is evaluated across six agent implementations, human browsing and conventional crawlers.", "topics": ["safety-security", "observability", "evaluation"], "updated_at": "2026-06-18", "url": "https://arxiv.org/abs/2606.20910"}
{"aliases": [], "authors": ["Iliana Fayolle", "Sihem Bouhenniche", "Samuel Pélissier", "Pierre Laperdrix", "Clémentine Maurice", "Walter Rudametkin"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "770c901e3490b2a43b02296136825e8e6bd035fb7494ea74b3a9af74be1afed1", "submitted_at": "2026-06-29T10:51:13Z", "updated_at": "2026-06-29T10:51:13Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2606.30119v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2606.30119", "identifiers": {"arxiv": "2606.30119"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:36Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2606.30119"}], "name": "On the Internet, Nobody Knows You're an LLM Bot: Unmasking Web Agents with Multi-Layer Fingerprinting", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-06-29", "relations": [], "review_status": "curated", "scope": "core", "summary": "Evaluates web agents against anti-bot mechanisms on instrumented honeysites and analyzes network, HTTP and browser fingerprints. The study compares resistance to blocking with the ability to distinguish agents, humans and agent implementations.", "topics": ["safety-security", "observability", "evaluation"], "updated_at": "2026-06-29", "url": "https://arxiv.org/abs/2606.30119"}
{"aliases": [], "authors": ["Wei Dong", "Tianyu Fu", "Zhe Yu", "Hanning Wang", "Anyang Su", "Zhizhou Fang", "Yuyang Chen", "Shuo Wang", "Minghui Wu", "Ping Jiang", "Zhen Lei", "Chenxu Zhao"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "cc6d846f43b9992388e94f3bcfdfc9a25d09b317fea62336aa22951c8fce9f8b", "submitted_at": "2026-07-07T10:27:31Z", "updated_at": "2026-07-07T10:27:31Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2607.06118v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2607.06118", "identifiers": {"arxiv": "2607.06118"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2607.06118"}], "name": "WebRetriever: A Large-Scale Comprehensive Benchmark for Efficient Web Agent Evaluation", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-07-07", "relations": [], "review_status": "curated", "scope": "core", "summary": "WebRetriever evaluates navigation, knowledge-assisted interaction and end-to-end information-extraction workflows under separate protocols. Its NavEval judge uses richer interaction context to diagnose cases where navigation success does not translate into useful task completion.", "topics": ["evaluation", "web-navigation", "observability"], "updated_at": "2026-07-07", "url": "https://arxiv.org/abs/2607.06118"}
{"aliases": [], "authors": ["Yong Liu", "Zhenyi Zhong", "Zhanpeng Shi"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "a779b2c1ff9ba3b531d399204b0986e3affcea81c67fdd7951e3b7da343b0e17", "submitted_at": "2026-07-16T00:35:25Z", "updated_at": "2026-07-16T00:35:25Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2607.14443v1"}}, "environments": ["desktop", "macos"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2607.14443", "identifiers": {"arxiv": "2607.14443"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2607.14443"}], "name": "Tactile: Giving Computer-Using Agents Hands and Feet", "notes": "The annotation was reviewed against the complete retained official abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2026-07-16", "relations": [], "review_status": "curated", "scope": "core", "summary": "Tactile turns accessibility, OCR and visual evidence into explicit desktop targets with executable actions and verification cues. Its execution loop chooses semantic actions or coordinate fallbacks while retaining provenance for replay and diagnosis.", "topics": ["grounding", "automation", "observability", "tool-integration", "robustness", "evaluation"], "updated_at": "2026-07-16", "url": "https://arxiv.org/abs/2607.14443"}
{"aliases": [], "authors": ["Madhumitha Venkatesan", "Shicheng Wen", "Jiajing Guo", "Jorge Piazentin Ono", "Liu Ren", "Dongyu Liu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "1179c657434761187f3fa3bed99f46530a69878578b7a74c68febae20e64d6b3", "submitted_at": "2026-07-16T16:48:39Z", "updated_at": "2026-07-16T16:48:39Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2607.15193v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2607.15193", "identifiers": {"arxiv": "2607.15193"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2607.15193"}], "name": "Plover: Steering GUI Agents through Plan-Centric Interaction", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-07-16", "relations": [], "review_status": "curated", "scope": "core", "summary": "Plover exposes GUI-agent plans and replanning as persistent artifacts that users can inspect and edit during execution. Its planner–executor design supports localized correction through plan edits, language guidance and screenshot-grounded interventions, studied through failure repair and workflow scenarios.", "topics": ["planning", "human-in-the-loop", "observability", "workflow"], "updated_at": "2026-07-16", "url": "https://arxiv.org/abs/2607.15193"}
{"aliases": [], "authors": ["Ruei-Che Chang", "Wenqian Xu", "Dingzeyu Li", "Bryan Wang", "Anhong Guo"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "2069acf693e0dced49d0e10e21bf9fe28b99f50f1b932837f16137e0d3170a80", "submitted_at": "2026-07-20T04:03:38Z", "updated_at": "2026-07-20T04:03:38Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2607.17527v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2607.17527", "identifiers": {"arxiv": "2607.17527"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2607.17527"}], "name": "Sidekick: Designing Communication for Effective Multitasking with Computer Use Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-07-20", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces Sidekick, a communication interface that uses ambient status cues, multimodal action summaries, and visual and verbal explanations of reasoning to help people monitor computer-use agents. A user study examines progress awareness, error traceability, and multitasking across background and foreground interaction.", "topics": ["human-in-the-loop", "observability", "long-horizon"], "updated_at": "2026-07-20", "url": "https://arxiv.org/abs/2607.17527"}
{"aliases": [], "authors": ["Zihan Dong", "Zhiyuan Ma", "Zekun Wang", "Yunqing Li", "Zirou Liu", "Ruixuan Deng", "Qishi Zhan", "Rui Qian"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "d75f6c942d6854a259556b7ef1060fa0f142755162da329580c6177cd1adef15", "submitted_at": "2026-07-30T15:29:42Z", "updated_at": "2026-07-30T15:29:42Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2607.28367v1"}}, "environments": ["web", "desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2607.28367", "identifiers": {"arxiv": "2607.28367"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2607.28367"}], "name": "How Benchmarks Mis-Score Computer-Use Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-07-30", "relations": [], "review_status": "curated", "scope": "core", "summary": "Audits failure-scored computer-use trajectories to distinguish agent errors from broken tasks and evaluator false negatives. It organizes reliability issues across task construction, trajectory evidence, scoring, and reporting, then derives corresponding evaluation design practices.", "topics": ["evaluation", "observability", "robustness"], "updated_at": "2026-07-30", "url": "https://arxiv.org/abs/2607.28367"}
{"aliases": [], "authors": ["Divake Kumar", "Sina Tayebati", "Devashri Naik", "Amanda Sofie Rios", "Nilesh Ahuja", "Omesh Tickoo", "Ranganath Krishnan", "Amit Ranjan Trivedi"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "14864aae8f977d474ff3e7dd3e38951374862abfb8077395f2dc5a9e1cc5a489", "submitted_at": "2026-08-28T00:50:47Z", "updated_at": "2026-08-28T00:50:47Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2608.27808v1"}}, "environments": ["desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2608.27808", "identifiers": {"arxiv": "2608.27808"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2608.27808"}], "name": "CURA: Certified Runtime Alarms for Computer-Use Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-08-28", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces CURA, a telemetry-based runtime monitor that turns computer-use trajectories into sequential failure alarms and can trigger mid-task oversight. Its certificate controls false alarms, while the study separately reports detection performance and limits of retrospective monitoring.", "topics": ["observability", "safety-security", "evaluation"], "updated_at": "2026-08-28", "url": "https://arxiv.org/abs/2608.27808"}
{"aliases": [], "authors": ["Sitong Pan", "Yipeng Shen", "Yilin Lu", "Caiwen Ding", "Lu Cheng", "Qianwen Wang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "967f1cdda69349da3280827c0b90301e4942c44398902ab6cfb868f1d3e94a84", "submitted_at": "2026-09-02T03:34:47Z", "updated_at": "2026-09-02T03:34:47Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2609.02057v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2609.02057", "identifiers": {"arxiv": "2609.02057"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2609.02057"}], "name": "Monitoring Web Agents Without Internal Signals: Observable Trajectories and Key-Step Supervision", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-09-02", "relations": [], "review_status": "curated", "scope": "core", "summary": "Predicts web-agent failure risk from observable trajectory prefixes using cross-step behavior and black-box consistency queries. Key-step labels preserve valid early prefixes of failed runs, and evaluation tests early intervention and transfer across website categories.", "topics": ["observability", "evaluation", "robustness", "interaction-history"], "updated_at": "2026-09-02", "url": "https://arxiv.org/abs/2609.02057"}
{"aliases": [], "authors": ["Zhilin Wang", "Shaokun Zhang", "Yifan Zhang", "Hao Zhang", "Jin Xu", "Binfeng Xu", "Jian Hu", "Yunheng Zou", "Karan Sapra", "Andrew Tao", "Jan Kautz", "Yi Dong"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "0d3cb920a677108910530788eb68c263127dba9510274f182910aed9c2f848c9", "submitted_at": "2026-09-21T16:55:24Z", "updated_at": "2026-09-21T16:55:24Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2609.24890v1"}}, "environments": ["desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2609.24890", "identifiers": {"arxiv": "2609.24890"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2609.24890"}], "name": "OSWorld-Pro: Process-based Evaluation for Computer Use Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-09-21", "relations": [], "review_status": "curated", "scope": "core", "summary": "Adds annotated intermediate subgoals to computer-use tasks and uses process-oriented evaluation to identify action and progress failures.", "topics": ["evaluation", "observability"], "updated_at": "2026-09-21", "url": "https://arxiv.org/abs/2609.24890"}
{"aliases": [], "authors": ["Shaoqing Zhang", "Kehai Chen", "Xuefeng Bai", "Zhuosheng Zhang", "Pengfei Zhang", "Yang Xiang", "Min Zhang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "2f4116455785579ece9a2ff1bae304b62f1cbe13a91db8d146bd281a27d19eb0", "submitted_at": "2026-09-28T01:48:48Z", "updated_at": "2026-09-29T09:05:42Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2609.34113v2"}}, "environments": ["android", "web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2609.34113", "identifiers": {"arxiv": "2609.34113"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2609.34113"}], "name": "GUITAR: Structured Failure Diagnosis of GUI Agents via State Transitions", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-09-28", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces GUITAR, which groups visually different screens into functional states and analyzes agent failures over a state-transition graph. It uses the resulting bottlenecks to diagnose recurring mobile and web errors and evaluate targeted guidance.", "topics": ["evaluation", "observability", "robustness"], "updated_at": "2026-09-29", "url": "https://arxiv.org/abs/2609.34113"}
{"aliases": [], "authors": ["Giulio Segalini", "Zhi Wen Soi", "Jérémie Decouchant", "Lydia Chen"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "945184e8fc790979a1f7e466ef6b4e727ddd11c3f285be0371fa4dda42615f40", "submitted_at": "2026-09-29T10:01:52Z", "updated_at": "2026-09-29T10:01:52Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2609.37176v1"}}, "environments": ["desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2609.37176", "identifiers": {"arxiv": "2609.37176"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2609.37176"}], "name": "Absorbed in Inertia: Activation Analysis for Computer-Use Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-09-29", "relations": [], "review_status": "curated", "scope": "core", "summary": "Studies repetitive, unproductive actions in desktop agents through their internal activation states. It proposes Reset, Reroute, Restore, a context intervention that temporarily resets an agent’s trajectory and later restores it to help escape repeated ineffective behavior.", "topics": ["robustness", "observability", "memory"], "updated_at": "2026-09-29", "url": "https://arxiv.org/abs/2609.37176"}
{"aliases": [], "authors": ["Geyi Yang", "Zikun Qu", "Xiang Li", "Zhiyong Wang", "Min Zhang", "Shipei Zeng", "Zhongxiang Dai"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "5c43224c865fba2a9f4c826a8e912ab0c07297716e4b7640bf312cce9c9c510e", "submitted_at": "2026-10-01T02:31:20Z", "updated_at": "2026-10-01T02:31:20Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.00948v1"}}, "environments": ["desktop", "windows"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2610.00948", "identifiers": {"arxiv": "2610.00948"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.00948"}], "name": "GUI-HARVEST: Self-Improving GUI Agents through Evidence-Driven Harness Evolution", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-10-01", "relations": [], "review_status": "curated", "scope": "core", "summary": "Improves GUI-agent execution code by analyzing action effects and repeated-task failures, then validating bounded changes with frozen model weights.", "topics": ["automation", "observability", "robustness"], "updated_at": "2026-10-01", "url": "https://arxiv.org/abs/2610.00948"}
{"aliases": [], "authors": ["Chengguang Gan", "Zimeng He", "Yoshihiro Tsujii", "Ken-ichiro Kobayashi", "Hiroki Itoh", "Kotaro Funakoshi"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "9a5b97264b3928c8dc74b955cc4baa7005cd34b0ce9e42cbc48a63bc6dd1f5a7", "submitted_at": "2026-10-01T11:31:37Z", "updated_at": "2026-10-01T11:31:37Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.01491v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2610.01491", "identifiers": {"arxiv": "2610.01491"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.01491"}], "name": "Auditing Web Agent Evaluation on WebArena-Lite: Human Review of Outcomes and Trajectories", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-10-01", "relations": [], "review_status": "curated", "scope": "core", "summary": "Audits WebArena-Lite outcomes and trajectories with human review to identify automatic-evaluator false negatives and the first consequential execution errors. It also studies explicit execution-state memory and procedural guidance, showing why final task scores can miss substantial differences in agent behavior.", "topics": ["evaluation", "observability", "memory"], "updated_at": "2026-10-01", "url": "https://arxiv.org/abs/2610.01491"}
{"aliases": [], "authors": ["Jiangang Han"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "c25f023b6ff66baa38661362d5730fc338e781db5bd23dd91216d3937ae730b0", "submitted_at": "2026-10-02T09:18:28Z", "updated_at": "2026-10-07T13:46:51Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2610.03036v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2610.03036", "identifiers": {"arxiv": "2610.03036"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.03036"}], "name": "WebFovea: When the Model Is Right but the Click Is Wrong -- Reliable Round Trips for Vision-Based Web Agents on Live Websites", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-10-02", "relations": [], "review_status": "curated", "scope": "core", "summary": "Describes WebFovea, a vision-based live-web agent whose execution loop checks action parsing, browser effects, feedback accuracy, and observation delivery. The report analyzes failures such as coordinate mismatches and silent control interactions and documents corresponding harness changes.", "topics": ["web-navigation", "robustness", "observability"], "updated_at": "2026-10-07", "url": "https://arxiv.org/abs/2610.03036"}
{"aliases": [], "authors": ["Michael Ofengenden", "Maksym Andriushchenko"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "03aea09f3ed1b75a8a6584efdfa51d45d1cfa9dc53fb3a25efd820c5b72e094a", "submitted_at": "2026-10-07T12:26:24Z", "updated_at": "2026-10-08T09:06:36Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2610.09944v2"}}, "environments": ["desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-20T13:48:10Z", "role": "primary", "sha256": "6d8c0463b6d50320246021f0394d0e39c9fd839f1ef1f742c8cdfa231872b059", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agent%22+OR+ti%3A%22browser+agent%22+OR+ti%3AGUI+OR+ti%3A%22Android+agent%22+OR+abs%3A%22computer-use+agents%22+OR+abs%3A%22GUI+agents%22+OR+abs%3A%22browser+agents%22+OR+abs%3A%22mobile+GUI%22%29+AND+submittedDate%3A%5B202610060000+TO+202610200000%5D&start=0&max_results=200&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T15:08:56.488063+00:00", "role": "primary", "sha256": "5424a5f9278ef276c84799a08d4e2e171ee93947a38c1d6670c36fa76dc84fe3", "supports": ["name", "kind", "summary", "topics", "scope", "notes"], "url": "https://arxiv.org/html/2610.09944v2"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T15:15:05.368243+00:00", "role": "primary", "sha256": "9bef76f95b41a7ca39c6fa3276d6122e4b27e1e0bd56a777edcfd7e11e7d7342", "supports": ["environments", "notes"], "url": "https://github.com/michaelofengenden/agenttimebench"}], "id": "arxiv:2610.09944", "identifiers": {"arxiv": "2610.09944"}, "kind": "paper", "last_verified_at": "2026-10-20T15:08:56.488063+00:00", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.09944"}], "name": "AgentTime: Can Agents Estimate and Control Their Own Runtime?", "notes": "This cross-domain suite includes 16 OSWorld 2.0 tasks among 222 tasks from 18 sources. A single run per task and requested duration, provider effects and incomplete native grades limit comparisons; elapsed time is not evidence of productive work. The OSWorld task 092 video judge was skipped without an API key. The inspected repository README identifies desktop-VM timing and states that the current 222-task forecasting runs will be added; its existing forecasting code covers an earlier study, and transcript work/wait labels are not public. No code or benchmark was executed and no license was verified.", "published_at": "2026-10-07", "relations": [], "review_status": "curated", "scope": "core", "summary": "Evaluates duration following, runtime forecasts and elapsed-time estimates in native agent harnesses, including OSWorld desktop tasks, with external timers, native graders and work-versus-wait analysis.", "topics": ["evaluation", "observability"], "updated_at": "2026-10-08", "url": "https://arxiv.org/abs/2610.09944"}
{"aliases": [], "authors": ["Hyobin Ahn", "Dongjae Lee"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "ff5470e792d14ced2cdff648d97e2606eea4b3c822046b7c10272fd2a5f1695f", "submitted_at": "2026-10-11T10:09:44Z", "updated_at": "2026-10-11T10:09:44Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.15502v1"}}, "environments": ["android", "web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-20T12:56:03.829575Z", "role": "primary", "sha256": "a8636ac679d2110181d4417d4aef040ddb24464597a7fa6fe6d2c35a2401a3b1", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22graphical+user+interface%22+OR+ti%3A%22screen+grounding%22+OR+ti%3A%22UI+grounding%22+OR+ti%3A%22desktop+agents%22+OR+ti%3A%22Android+agent%22+OR+ti%3A%22phone+agent%22+OR+ti%3A%22mobile+GUI%22+OR+ti%3A%22web+navigation%22+OR+ti%3A%22GUI+automation%22+OR+ti%3A%22browser+automation%22%29+AND+submittedDate%3A%5B202608210000+TO+202610201255%5D&start=0&max_results=200&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T13:12:58.069698Z", "role": "primary", "sha256": "a38b3ddc9d5d6744eefc937f86e174917f9b5de149323632d5d1b20283d79c8e", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://arxiv.org/html/2610.15502v1"}], "id": "arxiv:2610.15502", "identifiers": {"arxiv": "2610.15502"}, "kind": "paper", "last_verified_at": "2026-10-20T13:12:58.069698Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.15502"}], "name": "From Observations to Authorized Actions: Measuring Source Information in Android Agents", "notes": "The two prototypes use different workloads and trust assumptions on patched Android/WebView components. Approval is simulated; authenticated browser delivery and atomic capture remain unimplemented. Coarse source history can withhold benign work and persist across tasks. Direct-call checks and limited live-agent trials are separate, and writes are in-memory. No production vulnerability, general defense rate or released artifact is established.", "published_at": "2026-10-11", "relations": [], "review_status": "curated", "scope": "core", "summary": "Measures source-attribution limits in Android accessibility and separately tests application-history and selected-WebView-object checks for authorizing AppFunctions actions.", "topics": ["safety-security", "tool-integration", "observability"], "updated_at": "2026-10-11", "url": "https://arxiv.org/abs/2610.15502"}
{"aliases": [], "authors": ["Songping Wang", "Borui Li", "Qinglong Liu", "Yueming Lyu", "Yi Huang", "Fengting Li", "Chao Feng", "Caifeng Shan"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "ea753c1cb0666950d89f2781511f26872e3d5b1c2fd2434896248214160fa22c", "submitted_at": "2026-10-16T11:06:37Z", "updated_at": "2026-10-16T11:06:37Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.23303v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-20T05:52:01Z", "role": "primary", "sha256": "63e37cd26bfe69455a3fa835f3d6ba4cfcc5bb7dc91f90505920d37b1e08f0b6", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2610.19384%2C2610.22672%2C2610.22697%2C2610.23303%2C2610.23307%2C2610.24729&max_results=6"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T08:27:26.236797+00:00", "role": "primary", "sha256": "80bed80786f939973d0d13b4f892300fee02a6110e326247131fc83d257796de", "supports": ["name", "kind", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agent%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22+OR+ti%3A%22browser+agents%22+OR+ti%3A%22mobile+agent%22+OR+ti%3A%22mobile+agents%22+OR+ti%3A%22desktop+agent%22%29+AND+submittedDate%3A%5B202610130000+TO+202610192359%5D&start=0&max_results=200&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T10:42:09.967883+00:00", "role": "primary", "sha256": "ec9d647818be346ddbf7807be87b4789cd53e50c5d3d7a7904ad4f49bc71c38e", "supports": ["summary", "environments", "topics", "scope", "notes"], "url": "https://arxiv.org/html/2610.23303v1"}], "id": "arxiv:2610.23303", "identifiers": {"arxiv": "2610.23303"}, "kind": "paper", "last_verified_at": "2026-10-20T08:27:26.236797+00:00", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.23303"}], "name": "MobileJailBench: Red Queen Jailbreak Benchmarking for Mobile Agents", "notes": "Reviewed the complete retained abstract and selected exact-v1 threat-model, construction and evaluation sections. Execution uses a browser-based Android-like simulator, not real Android isolation or live services. The attacker controls instructions and designated carriers, not privileged simulator state/APIs or system/model/harness internals. Rates depend on release, budget and adjudication coverage. SAF labels track observable expressions and actions, not latent reasoning; unresolved episodes are excluded from rate denominators. No independent reproduction.", "published_at": "2026-10-16", "relations": [], "review_status": "curated", "scope": "core", "summary": "MobileJailBench evaluates mobile-agent jailbreaks in a browser-based Android-like simulator, combining feedback-driven attack evolution, separately admitted task expansions and evidence-grounded outcome adjudication.", "topics": ["safety-security", "mobile-navigation", "evaluation", "observability", "cross-app"], "updated_at": "2026-10-16", "url": "https://arxiv.org/abs/2610.23303"}
{"aliases": [], "authors": ["Songping Wang", "Peiyang Li", "Yi Huang", "Fengting Li", "Xuhao Ran", "Yueming Lyu", "Chao Feng", "Caifeng Shan"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "3fc22fbeda19ac2947b00af73801f0674c550e0d7c78919f811cae754b06128c", "submitted_at": "2026-10-16T11:10:30Z", "updated_at": "2026-10-16T11:10:30Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.23307v1"}}, "environments": ["android"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-20T05:52:01Z", "role": "primary", "sha256": "63e37cd26bfe69455a3fa835f3d6ba4cfcc5bb7dc91f90505920d37b1e08f0b6", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2610.19384%2C2610.22672%2C2610.22697%2C2610.23303%2C2610.23307%2C2610.24729&max_results=6"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T08:27:26.236797+00:00", "role": "primary", "sha256": "80bed80786f939973d0d13b4f892300fee02a6110e326247131fc83d257796de", "supports": ["name", "kind", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agent%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22+OR+ti%3A%22browser+agents%22+OR+ti%3A%22mobile+agent%22+OR+ti%3A%22mobile+agents%22+OR+ti%3A%22desktop+agent%22%29+AND+submittedDate%3A%5B202610130000+TO+202610192359%5D&start=0&max_results=200&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T09:17:36.223826+00:00", "role": "primary", "sha256": "3803984a185b6258276b87b87ac1991583878fb41dc22618ff743f6f69c7bc7e", "supports": ["summary", "environments", "topics", "scope", "notes"], "url": "https://arxiv.org/html/2610.23307v1"}], "id": "arxiv:2610.23307", "identifiers": {"arxiv": "2610.23307"}, "kind": "paper", "last_verified_at": "2026-10-20T08:27:26.236797+00:00", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.23307"}], "name": "MobileInject: Evaluating Reasoning-to-Action Security of MLLM-Based Mobile Agents under Cross-Modal Consistency Injection", "notes": "Reviewed the complete abstract and selected v1 construction, threat-model and evaluation sections. The attacker modifies observation surfaces, not system instructions, model parameters or the OS. State oracles are distinct from emitted-rationale diagnoses, whose labels include outcome-dependent constraints. Those diagnostics are not independent evidence of hidden-reasoning causality. No reproduction or release verification.", "published_at": "2026-10-16", "relations": [], "review_status": "curated", "scope": "core", "summary": "MobileInject evaluates visual/textual observation injection in stateful Android workflows using resettable emulators, benign/adversarial state oracles and trace-based diagnosis of emitted rationales and actions.", "topics": ["safety-security", "mobile-navigation", "evaluation", "observability", "cross-app"], "updated_at": "2026-10-16", "url": "https://arxiv.org/abs/2610.23307"}
{"aliases": [], "authors": ["Ananya Gubbi Mohanbabu", "Rosiana Natalie", "Anhong Guo", "Amy Pavel"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "9858b5ec15ee953504bd00e26b455c1e81104f80429c2dd112de17f9d23977ab", "submitted_at": "2026-10-19T16:39:11Z", "updated_at": "2026-10-19T16:39:11Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.26703v1"}}, "environments": ["desktop", "web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-20T05:52:00Z", "role": "primary", "sha256": "af310824eea97aac8209023ff53bad7bea77778816586b10bb94110148347af9", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agent%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B202610060000+TO+202610200000%5D&start=0&max_results=200&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T08:27:26.236797+00:00", "role": "primary", "sha256": "80bed80786f939973d0d13b4f892300fee02a6110e326247131fc83d257796de", "supports": ["name", "kind", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agent%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22+OR+ti%3A%22browser+agents%22+OR+ti%3A%22mobile+agent%22+OR+ti%3A%22mobile+agents%22+OR+ti%3A%22desktop+agent%22%29+AND+submittedDate%3A%5B202610130000+TO+202610192359%5D&start=0&max_results=200&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T09:17:30.049627+00:00", "role": "primary", "sha256": "9a964cbf50123887ad20bf156e69c0bbe4567d6b2c28e4cf202d5538eab8294c", "supports": ["summary", "environments", "topics", "scope", "notes"], "url": "https://arxiv.org/html/2610.26703v1"}], "id": "arxiv:2610.26703", "identifiers": {"arxiv": "2610.26703"}, "kind": "paper", "last_verified_at": "2026-10-20T08:27:26.236797+00:00", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.26703"}], "name": "Comparing Computer Use Behavior Across Multiple Levels of Abstraction", "notes": "Reviewed the complete abstract and selected v1 taxonomy, annotation, evaluation and limitations sections. MAP analyzes existing human/agent traces rather than deploying a new acting policy. Annotation and timing coverage depend on the source datasets; manual agreement reflects a particular labeling scheme. Behavioral findings do not establish a privacy guarantee. No independent reproduction.", "published_at": "2026-10-19", "relations": [], "review_status": "curated", "scope": "core", "summary": "MAP annotates computer-use traces at linked operation, function/consequence, method and goal levels, then compares human and agent workflows through sequence, pattern and effort metrics.", "topics": ["evaluation", "observability"], "updated_at": "2026-10-19", "url": "https://arxiv.org/abs/2610.26703"}
{"aliases": [], "authors": [], "details": {}, "environments": ["web"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:05Z", "role": "primary", "sha256": "86d7fbe30596de4a1f1a1df2a986f5f247351a91c0c593dc4c29ba9c70d28c8f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://raw.githubusercontent.com/aaronlab/browsertrace/HEAD/README.md"}, {"method": "license_header", "retrieved_at": "2026-10-16T06:42:06Z", "role": "primary", "sha256": "7b4ea96686420061c9d0e1ad4984d7c5efb7e1a62ce6dffd47862f02a7260867", "supports": ["licenses.code"], "url": "https://raw.githubusercontent.com/aaronlab/browsertrace/HEAD/LICENSE"}], "id": "github:aaronlab/browsertrace", "identifiers": {}, "kind": "tool", "last_verified_at": "2026-10-16T06:36:05Z", "licenses": {"code": {"basis": "repository LICENSE header; not a weights/data license", "declared": "MIT", "retrieved_at": "2026-10-16T06:42:06Z", "url": "https://raw.githubusercontent.com/aaronlab/browsertrace/HEAD/LICENSE"}, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/aaronlab/browsertrace"}], "name": "BrowserTrace", "notes": "", "published_at": null, "relations": [], "review_status": "curated", "scope": "core", "summary": "Local, cloud-free replay debugger that records each Browser Use step's screenshot, URL, action and model I/O as a timeline and reports the step at which a failed run diverges from a known-good one.", "topics": ["observability"], "updated_at": null, "url": "https://github.com/aaronlab/browsertrace"}
{"aliases": [], "authors": [], "details": {}, "environments": ["desktop"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-16T10:42:12Z", "role": "primary", "sha256": "2e42f01578710d91145412f7d1ecb93979db96c698f677915284da1a1d592588", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://raw.githubusercontent.com/GaryYang12345/GUI-HARVEST/HEAD/README.md"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "license_header", "retrieved_at": "2026-10-16T10:48:42Z", "role": "primary", "sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30", "supports": ["licenses.code"], "url": "https://raw.githubusercontent.com/GaryYang12345/GUI-HARVEST/HEAD/LICENSE"}], "id": "github:garyyang12345/gui-harvest", "identifiers": {}, "kind": "framework", "last_verified_at": "2026-10-16T10:42:12Z", "licenses": {"code": {"basis": "repository LICENSE header; not a weights/data license", "declared": "Apache-2.0", "retrieved_at": "2026-10-16T10:48:42Z", "url": "https://raw.githubusercontent.com/GaryYang12345/GUI-HARVEST/HEAD/LICENSE"}, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/GaryYang12345/GUI-HARVEST"}], "name": "GUI-HARVEST", "notes": "Requires a separately installed benchmark and seed agent harness; benchmark tasks and model weights remain fixed during its optimization procedure.", "published_at": null, "relations": [{"evidence_url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending", "target_id": "arxiv:2610.00948", "type": "documented_by"}], "review_status": "curated", "scope": "core", "summary": "Optimizes GUI-agent execution code from repeated rollout evidence and checks proposed changes against fixed validation gates.", "topics": ["automation", "observability", "robustness"], "updated_at": null, "url": "https://github.com/GaryYang12345/GUI-HARVEST"}
{"aliases": [], "authors": [], "details": {"availability": "Public browser-agent source and npm release metadata.", "license_review": {"code": {"evidence_urls": ["https://raw.githubusercontent.com/omxyz/lumen/HEAD/LICENSE"], "reason": "Publisher project-source license; external model and browser-service terms are separate."}}, "package_version": "0.2.0", "repository_revision": "b1ad26a0784645ac3a97d402db99cd5d17f86334"}, "environments": ["web"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-17T12:30:55Z", "role": "primary", "sha256": "7708f434142498959bb5ed5de35ac1b945f9ad6e070e436b4e73819b7db1df7f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://raw.githubusercontent.com/omxyz/lumen/HEAD/README.md"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T12:30:55Z", "role": "primary", "sha256": "32b02ea43d19929d82e8b8c52bb8e448553aebdd96a8754d144927cb0f34c5b6", "supports": ["identity"], "url": "https://api.github.com/repos/omxyz/lumen"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T12:32:05Z", "role": "primary", "sha256": "9911953e501a245d657f3835b22c6822ad7d990b96f03e3cba93a15ebe4f70a6", "supports": ["release_status", "details.repository_revision"], "url": "https://api.github.com/repos/omxyz/lumen/git/trees/main?recursive=1"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T12:34:10Z", "role": "primary", "sha256": "34a03b6dd8cd1960d5592a6569e2a1c1f6c1b2697da1386a58d0ca64157f059e", "supports": ["identity", "release_status", "details.package_version", "notes"], "url": "https://registry.npmjs.org/@omxyz%2Flumen/latest"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T12:34:10Z", "role": "primary", "sha256": "30dd5336a8a45b548b9051e751d45923a9864b3810a36c836df9b07447ca3c5c", "supports": ["licenses.code"], "url": "https://raw.githubusercontent.com/omxyz/lumen/HEAD/LICENSE"}], "id": "github:omxyz/lumen", "identifiers": {}, "kind": "agent", "last_verified_at": "2026-10-17T12:30:55Z", "licenses": {"code": {"basis": "Publisher project-source license; external model and browser-service terms are separate.", "declared": "MIT", "retrieved_at": "2026-10-17T12:34:10Z", "url": "https://raw.githubusercontent.com/omxyz/lumen/HEAD/LICENSE"}, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/omxyz/lumen"}], "name": "Lumen browser agent", "notes": "Public TypeScript source and npm package metadata are available; the inspected package version is 0.2.0. The package declares Node ^20.19.0 or >=22.12.0 and requires a browser and model provider. Its README benchmark is explicitly preliminary and uses a small WebVoyager subset; this catalog makes no comparative performance or independent safety claim. Referenced research influences are not asserted as papers introducing Lumen.", "published_at": null, "relations": [], "review_status": "curated", "scope": "core", "summary": "A screenshot-driven browser agent over Chrome DevTools Protocol, with resumable sessions, context compression, action replay and configurable completion checks.", "topics": ["web-navigation", "memory", "automation", "observability"], "updated_at": null, "url": "https://github.com/omxyz/lumen"}
{"aliases": [], "authors": [], "details": {"license_review": {"code": {"evidence_urls": ["https://raw.githubusercontent.com/wlanderson0/orbit/588b3035f92450ec4f0496bb8b629daa205e84c7/LICENSE", "https://raw.githubusercontent.com/wlanderson0/orbit/588b3035f92450ec4f0496bb8b629daa205e84c7/NOTICE"], "reason": "Record only the publisher's top-level Apache-2.0 declaration for Orbit-original code. NOTICE explicitly preserves per-component terms; neither bundled or external data nor model weights inherit this declaration. Its account of BrowserART licensing is not an independent resolution of upstream conflicting declarations."}}}, "environments": ["web", "desktop"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-20T13:41:22.467183+00:00", "role": "primary", "sha256": "b6743e3e9e5256d7afa8bb411baf246b60ba71bcfb84eda5a816fba3ddd7d78b", "supports": ["name", "kind", "summary", "topics", "scope", "notes"], "url": "https://raw.githubusercontent.com/wlanderson0/orbit/588b3035f92450ec4f0496bb8b629daa205e84c7/README.md"}, {"method": "github_repository_identity", "retrieved_at": "2026-10-20T13:39:57.052202+00:00", "role": "primary", "sha256": "0436414b9bcc4307f3bcde476a544da6fe1dd5baa90939eed392942490fbfbf7", "supports": ["id", "url"], "url": "https://api.github.com/repos/wlanderson0/orbit"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T13:42:37.098639+00:00", "role": "primary", "sha256": "c47d046b129999da5e63b7723a3f3e2e0cc31c27ce74b45f71f8c6d0df1fc0d8", "supports": ["summary", "environments", "scope", "notes"], "url": "https://raw.githubusercontent.com/wlanderson0/orbit/588b3035f92450ec4f0496bb8b629daa205e84c7/orbit/scenarios/browser/browserart/tools.py"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T14:10:32.910533Z", "role": "primary", "sha256": "2cce76b22252a415d64c356b2062771e528b554dd83b6f88fca769f48fa8a826", "supports": ["summary", "environments", "scope", "notes"], "url": "https://raw.githubusercontent.com/wlanderson0/orbit/588b3035f92450ec4f0496bb8b629daa205e84c7/orbit/scenarios/desktop/osworld/tools.py"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T14:09:35.060985Z", "role": "primary", "sha256": "66515d7ee87eab5a2f42c26132d0bf3e2a3b9c006156d2378181cf91966a2903", "supports": ["notes"], "url": "https://raw.githubusercontent.com/wlanderson0/orbit/588b3035f92450ec4f0496bb8b629daa205e84c7/docs/benchmark-qualification.md"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T14:10:29.541892Z", "role": "primary", "sha256": "94c781028dc80b2a748d759fae2456f6a986da7487651a6bcbb179e8ad4fe023", "supports": ["notes"], "url": "https://raw.githubusercontent.com/wlanderson0/orbit/588b3035f92450ec4f0496bb8b629daa205e84c7/orbit/scenarios/browser/browserart/scorer.py"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T13:41:29.036076+00:00", "role": "primary", "sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30", "supports": ["licenses.code"], "url": "https://raw.githubusercontent.com/wlanderson0/orbit/588b3035f92450ec4f0496bb8b629daa205e84c7/LICENSE"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T14:09:31.550310Z", "role": "primary", "sha256": "f68a64f775b1f329416f2fe294bb07b5e475893f77d1cdb5300bdb08acf95410", "supports": ["licenses.code", "licenses.data", "notes"], "url": "https://raw.githubusercontent.com/wlanderson0/orbit/588b3035f92450ec4f0496bb8b629daa205e84c7/NOTICE"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T13:12:28.833839+00:00", "role": "primary", "sha256": "3ea667f4187ba601011e5ff00c4d63676b6d9903d12b80a0d7d5dd9920f00c65", "supports": ["relations"], "url": "https://arxiv.org/html/2609.33102v1"}], "id": "github:wlanderson0/orbit", "identifiers": {}, "kind": "framework", "last_verified_at": "2026-10-20T13:41:22.467183+00:00", "licenses": {"code": {"basis": "Complete root Apache License 2.0 text for Orbit-original code. NOTICE preserves separate third-party component terms, including MIT-vendored modules; this is not a uniform license for the entire dependency/data bundle.", "declared": "Apache-2.0", "retrieved_at": "2026-10-20T13:41:29.036076+00:00", "url": "https://raw.githubusercontent.com/wlanderson0/orbit/588b3035f92450ec4f0496bb8b629daa205e84c7/LICENSE"}, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/wlanderson0/orbit"}], "name": "ORBIT framework", "notes": "Reviewed version 1.0.3 source at commit 588b3035, not an asserted reproduction of the paper. BrowserART tools return accessibility trees, including the screenshot-named refresh. Qualification is scenario-specific: the project documents browser-judge and desktop-scoring limits; its opt-in exact-send oracle covers one derived local task, not SMTP delivery. Runtime profiles and scoring policies are separate. Apache-2.0 is the top-level Orbit code declaration; third-party components and external data/environments retain separate terms. No installation, release archive or live evaluation was tested.", "published_at": null, "relations": [{"evidence_url": "https://arxiv.org/html/2609.33102v1", "target_id": "arxiv:2609.33102", "type": "documented_by"}], "review_status": "curated", "scope": "core", "summary": "Inspect-based multi-agent evaluation framework with role, scheduling and communication controls, BrowserART accessibility-tree actions, and desktop mouse, keyboard and screenshot tools.", "topics": ["evaluation", "safety-security", "web-navigation", "observability"], "updated_at": null, "url": "https://github.com/wlanderson0/orbit"}