{
  "version": 1,
  "reviewed_only": true,
  "updated_at": "2026-11-03T02:01:03Z",
  "weeks": [
    "2026-W45",
    "2026-W44"
  ],
  "dates": [
    "2026-11-02",
    "2026-10-30",
    "2026-10-28"
  ],
  "recaps": [
    {
      "period": "2026-10-14--2026-10-27",
      "kind": "retrospective",
      "start": "2026-10-14",
      "end": "2026-10-27",
      "title_zh": "创刊回顾 · 10 月 14–27 日",
      "title_en": "Inaugural retrospective · October 14–27"
    }
  ],
  "items": [
    {
      "id": "arxiv:2610.42141",
      "arxiv_id": "2610.42141",
      "version": 1,
      "title": "RoboArchitect: Automating the Design of Robotic Agents with Persistent Memory",
      "short_title": "RoboArchitect",
      "authors": [
        "Shengguang Wu",
        "Suning Huang",
        "Cherie Ho",
        "Ria Doshi",
        "Hao Zhu",
        "Serena Yeung-Levy"
      ],
      "published": "2026-10-30T17:58:48Z",
      "updated": "2026-10-30T17:58:48Z",
      "hf_daily_dates": [
        "2026-11-02"
      ],
      "announced": "2026-10-30",
      "announced_source": "arxiv",
      "news_date": "2026-11-02",
      "window_status": "in_window",
      "in_repo": false,
      "gate": {},
      "direction": {
        "primary": "harness",
        "harness_sub": "evolve_main",
        "secondary": [
          "embodied"
        ]
      },
      "summary_zh": "在仿真中用轨迹反馈修订冻结策略外的控制代码，并在留出任务评估。",
      "summary_en": "Uses rollout feedback to revise control code around a frozen robotic policy, evaluating retained harnesses on held-out tasks in simulation.",
      "evidence_quick": {
        "held_out": {
          "value": "unknown",
          "quote": null
        },
        "matched_budget": {
          "value": "unknown",
          "quote": null
        },
        "cross_model": {
          "value": "unknown",
          "quote": null
        },
        "code_released": {
          "value": "unknown",
          "quote": null
        }
      },
      "evidence_notice": "Automated, unverified",
      "links": {
        "hf": "https://huggingface.co/papers/2610.42141",
        "arxiv": "https://arxiv.org/abs/2610.42141",
        "code": "https://github.com/roboarchitect-ai/RoboArchitect"
      },
      "source_hits": [
        "hf-daily"
      ],
      "hf_upvotes": 4,
      "week": "2026-W45",
      "review_decision": "accept",
      "repo_id": null,
      "review_date": "2026-11-02",
      "edition_kind": "daily",
      "daily_section": "hf_today"
    },
    {
      "id": "arxiv:2610.36077",
      "arxiv_id": "2610.36077",
      "version": 1,
      "title": "The Trinity:Capability-Driven Recursive Self-Improvement for Multimodal Agents",
      "short_title": "The Trinity",
      "authors": [
        "Weidong Tang",
        "Ruiheng Wang",
        "Jingxuan Zhang",
        "Hexin Wang",
        "Haoyuan Chen",
        "Zhengfeiyang Zhang",
        "Yixing Ma",
        "Zichen Jiang",
        "Ziyu Guo",
        "Haodong Li",
        "Zhengbo Jiao",
        "Zhiqing Cui",
        "Xinxiang Yin",
        "Shuo Liang",
        "Zhenglin Wan",
        "Hongzhan Lin",
        "Wangbo Zhao Kaipeng Zhang",
        "Pengfei Zhou",
        "Odin Zhang",
        "Yang You"
      ],
      "published": "2026-10-26T18:03:34Z",
      "updated": "2026-10-26T18:03:34Z",
      "hf_daily_dates": [
        "2026-10-28"
      ],
      "announced": "2026-10-26",
      "announced_source": "arxiv",
      "news_date": "2026-10-28",
      "window_status": "in_window",
      "in_repo": false,
      "gate": {
        "harness_specific": "yes",
        "improved_object": "skill_files",
        "improved_object_zh": "技能文件（能力模块）",
        "verdict": "harness"
      },
      "direction": {
        "primary": "harness",
        "harness_sub": "evolve_main",
        "secondary": [
          "data_task"
        ]
      },
      "summary_zh": "按执行反馈修订冻结模型外的协议模块并沿固定能力树继承；评测技能另行准备。",
      "summary_en": "Revises protocol modules around a frozen model using execution feedback and inherits them along a fixed capability tree; evaluation skills are prepared separately.",
      "evidence_quick": {
        "held_out": {
          "value": "yes",
          "quote": "26 points across 18 unseen benchmarks"
        },
        "matched_budget": {
          "value": "unknown",
          "quote": null
        },
        "cross_model": {
          "value": "unknown",
          "quote": null
        },
        "code_released": {
          "quote": "https://github.com/InfRec-Cardinal/The-Trinity",
          "value": "yes"
        }
      },
      "evidence_notice": "Automated, unverified",
      "links": {
        "hf": "https://huggingface.co/papers/2610.36077",
        "arxiv": "https://arxiv.org/abs/2610.36077",
        "code": "https://github.com/InfRec-Cardinal/The-Trinity"
      },
      "source_hits": [
        "arxiv-harness",
        "arxiv-rsi",
        "arxiv-version-check",
        "hf-daily"
      ],
      "hf_upvotes": 540,
      "week": "2026-W44",
      "review_decision": "accept",
      "repo_id": null,
      "review_date": "2026-11-02",
      "edition_kind": "daily",
      "daily_section": "hf_today"
    },
    {
      "id": "arxiv:2610.35434",
      "arxiv_id": "2610.35434",
      "version": 1,
      "title": "Skill^{infty}: Recursive Self-Improvement for Agent Skill Optimization",
      "short_title": "Skill^∞",
      "authors": [
        "Wangcheng Tao",
        "Zhanglu Yan",
        "Han Wu",
        "Weng-Fai Wong"
      ],
      "published": "2026-10-26T00:00:00.000Z",
      "updated": "2026-10-26T00:00:00.000Z",
      "hf_daily_dates": [
        "2026-10-27"
      ],
      "announced": "2026-10-26",
      "announced_source": "published",
      "news_date": "2026-10-27",
      "window_status": "backlog",
      "in_repo": false,
      "gate": {
        "verdict": "harness",
        "harness_specific": "yes",
        "improved_object_zh": "元技能与任务技能"
      },
      "direction": {
        "primary": "harness",
        "harness_sub": "improver",
        "secondary": [
          "meta"
        ]
      },
      "summary_zh": "元技能用同一流程优化任务技能并优化自身",
      "summary_en": "A meta-skill uses the same optimization loop to improve task skills and itself.",
      "evidence_quick": {
        "code_released": {
          "value": "unknown",
          "quote": null
        },
        "cross_model": {
          "quote": null,
          "value": "unknown"
        },
        "held_out": {
          "quote": null,
          "value": "unknown"
        },
        "matched_budget": {
          "quote": null,
          "value": "unknown"
        }
      },
      "evidence_notice": "Automated, unverified",
      "links": {
        "hf": "https://huggingface.co/papers/2610.35434",
        "arxiv": "https://arxiv.org/abs/2610.35434"
      },
      "source_hits": [
        "arxiv-harness",
        "arxiv-rsi",
        "hf-daily"
      ],
      "hf_upvotes": 2,
      "week": "2026-W44",
      "review_decision": "accept",
      "repo_id": null,
      "review_date": "2026-10-27",
      "daily_section": "historical",
      "edition_kind": "retrospective",
      "recap_period": "2026-10-14--2026-10-27"
    },
    {
      "id": "arxiv:2610.34722",
      "arxiv_id": "2610.34722",
      "version": 1,
      "title": "CEPO: A General Framework for Improving Prompt Optimization in Multi-Agent Systems",
      "short_title": "CEPO",
      "authors": [
        "Jongha Kim",
        "Seohyun Lee",
        "Jonathan Minho Chung",
        "Jinsung Yoon",
        "Hyunwoo J. Kim"
      ],
      "published": "2026-10-26",
      "updated": "2026-10-26",
      "hf_daily_dates": [
        "2026-11-02"
      ],
      "announced": "2026-10-26",
      "announced_source": "published",
      "news_date": "2026-11-02",
      "window_status": "in_window",
      "in_repo": false,
      "gate": {
        "harness_specific": "yes",
        "improved_object": "system_prompt",
        "improved_object_zh": "系统提示",
        "verdict": "borderline"
      },
      "direction": {
        "primary": "harness",
        "harness_sub": "component",
        "secondary": []
      },
      "summary_zh": "用执行反馈分别修订协调者与执行器提示，经验证集选择后用于后续任务。",
      "summary_en": "Uses execution feedback to revise coordinator and executor prompts separately, selecting prompts on validation tasks for subsequent use.",
      "evidence_quick": {
        "held_out": {
          "value": "unknown",
          "quote": null
        },
        "matched_budget": {
          "value": "unknown",
          "quote": null
        },
        "cross_model": {
          "value": "unknown",
          "quote": null
        },
        "code_released": {
          "quote": null,
          "value": "unknown"
        }
      },
      "evidence_notice": "Automated, unverified",
      "links": {
        "arxiv": "https://arxiv.org/abs/2610.34722",
        "code": "https://github.com/mlvlab/CEPO",
        "s2": "https://www.semanticscholar.org/paper/02bfc41201571241bed2f14d859e7c09ba7eddc6"
      },
      "source_hits": [
        "arxiv-version-check",
        "hf-daily",
        "s2-citations"
      ],
      "hf_upvotes": 15,
      "week": "2026-W45",
      "review_decision": "accept",
      "repo_id": null,
      "review_date": "2026-11-02",
      "edition_kind": "daily",
      "daily_section": "hf_today"
    },
    {
      "id": "arxiv:2610.33106",
      "arxiv_id": "2610.33106",
      "version": 1,
      "title": "VibeGame: Prompt-to-Game Development with AI-Native Engine and Self-Evolving Adversarial Agent Team",
      "short_title": "VibeGame",
      "authors": [
        "Wenbo Hu",
        "Ken Li",
        "Jiazhe Wei",
        "Yukang Cao",
        "Weiyi Hong",
        "Jiayi Dai",
        "Chenjun Bai",
        "Jiajun Liang",
        "Yucheng Liao",
        "Ruichuan An",
        "Zeyu Lou",
        "Haofan Wang",
        "Yueming Lyu",
        "Jingkang Yang",
        "Ziwei Liu",
        "Chenyang Si"
      ],
      "published": "2026-10-22T18:03:17Z",
      "updated": "2026-10-22T18:03:17Z",
      "hf_daily_dates": [
        "2026-10-30"
      ],
      "announced": "2026-10-22",
      "announced_source": "arxiv",
      "news_date": "2026-10-30",
      "window_status": "in_window",
      "in_repo": false,
      "gate": {
        "harness_specific": "yes",
        "improved_object": "skill_files",
        "improved_object_zh": "技能文件/模块库",
        "verdict": "harness"
      },
      "direction": {
        "primary": "harness",
        "harness_sub": "component",
        "secondary": []
      },
      "summary_zh": "从已验证项目提炼可复用协作契约供后续开发；自进化仅在经验消融中评估。",
      "summary_en": "Distills reusable collaboration contracts from validated projects for later development; self-evolution is evaluated separately through an experience-bundle ablation.",
      "evidence_quick": {
        "held_out": {
          "value": "unknown",
          "quote": null
        },
        "matched_budget": {
          "value": "unknown",
          "quote": null
        },
        "cross_model": {
          "value": "unknown",
          "quote": null
        },
        "code_released": {
          "quote": null,
          "value": "unknown"
        }
      },
      "evidence_notice": "Automated, unverified",
      "links": {
        "hf": "https://huggingface.co/papers/2610.33106",
        "arxiv": "https://arxiv.org/abs/2610.33106",
        "code": "https://github.com/tettethu/VibeGame"
      },
      "source_hits": [
        "arxiv",
        "arxiv-harness",
        "arxiv-version-check",
        "hf-daily"
      ],
      "hf_upvotes": 22,
      "week": "2026-W44",
      "review_decision": "accept",
      "repo_id": null,
      "review_date": "2026-11-02",
      "edition_kind": "daily",
      "daily_section": "hf_today"
    },
    {
      "id": "arxiv:2610.29560",
      "arxiv_id": "2610.29560",
      "version": 1,
      "title": "From VLM Planners to Robotic Agents: Self-Evolving Harnesses for History-Dependent Manipulation",
      "short_title": "SERA",
      "authors": [
        "Tingwen Du",
        "Haoyu Guo",
        "Hanxue Zhang",
        "Xuanxu Wang",
        "Junting Dong",
        "Chao Gao",
        "Mingfei Han",
        "Tao Lu",
        "Mulin Yu",
        "Chunhua Shen",
        "Xiaojun Chang",
        "Zhihui Li"
      ],
      "published": "2026-10-21T00:00:00.000Z",
      "updated": "2026-10-21T00:00:00.000Z",
      "hf_daily_dates": [
        "2026-10-22"
      ],
      "announced": "2026-10-21",
      "announced_source": "published",
      "news_date": "2026-10-22",
      "window_status": "backlog",
      "in_repo": false,
      "gate": {
        "verdict": "harness",
        "harness_specific": "yes",
        "improved_object_zh": "VLM 规划器的 Harness"
      },
      "direction": {
        "primary": "harness",
        "harness_sub": "evolve_main",
        "secondary": [
          "embodied"
        ]
      },
      "summary_zh": "固定 VLA 执行器，可编辑 Harness 依执行经验自进化",
      "summary_en": "The VLM planner evolves its editable Harness from execution experience while the VLA executor remains fixed.",
      "evidence_quick": {
        "code_released": {
          "value": "yes",
          "quote": "https://github.com/StevenKKXS/SERA"
        },
        "cross_model": {
          "quote": null,
          "value": "unknown"
        },
        "held_out": {
          "quote": "fixed before held-out testing",
          "value": "yes"
        },
        "matched_budget": {
          "quote": null,
          "value": "unknown"
        }
      },
      "evidence_notice": "Automated, unverified",
      "links": {
        "hf": "https://huggingface.co/papers/2610.29560",
        "arxiv": "https://arxiv.org/abs/2610.29560",
        "code": "https://github.com/StevenKKXS/SERA"
      },
      "source_hits": [
        "arxiv-harness",
        "hf-daily"
      ],
      "hf_upvotes": 33,
      "week": "2026-W44",
      "review_decision": "accept",
      "repo_id": null,
      "review_date": "2026-10-27",
      "daily_section": "historical",
      "edition_kind": "retrospective",
      "recap_period": "2026-10-14--2026-10-27"
    },
    {
      "id": "arxiv:2610.28601",
      "arxiv_id": "2610.28601",
      "version": 1,
      "title": "Evolving Agent Interfaces",
      "short_title": "Evolving Agent Interfaces (Agent S4)",
      "authors": [
        "Saaket Agashe",
        "Yuzhe Yang",
        "Gonzalo Gonzalez-Pumariega",
        "Bhuvan Chandra Koduru",
        "Chih-Lun Li",
        "Jiachen Yang",
        "Ang Li",
        "Xin Eric Wang"
      ],
      "published": "2026-10-20T00:00:00.000Z",
      "updated": "2026-10-20T00:00:00.000Z",
      "hf_daily_dates": [
        "2026-10-21"
      ],
      "announced": "2026-10-20",
      "announced_source": "published",
      "news_date": "2026-10-21",
      "window_status": "backlog",
      "in_repo": true,
      "gate": {
        "verdict": "harness",
        "harness_specific": "yes",
        "improved_object": "none",
        "improved_object_zh": "详见证据卡"
      },
      "direction": {
        "primary": "harness",
        "harness_sub": "component",
        "secondary": []
      },
      "summary_zh": "元 Agent 阅读合成任务上的无成功标签轨迹，写入并修订接口仓库；种群选择按元 Agent 看不到的验证任务决定保留哪些仓库。",
      "summary_en": "A meta agent reads trajectories without success labels on synthetic tasks, writing and revising interface repositories; population selection retains repositories using validation tasks hidden from the meta agent.",
      "evidence_quick": {
        "held_out": {
          "value": "yes",
          "quote": "papers/agent-s4.md"
        },
        "matched_budget": {
          "value": "unknown",
          "quote": null
        },
        "cross_model": {
          "value": "unknown",
          "quote": null
        },
        "code_released": {
          "value": "yes",
          "quote": "papers/agent-s4.md"
        }
      },
      "evidence_notice": "Curated evidence; see evidence card",
      "links": {
        "hf": "https://huggingface.co/papers/2610.28601",
        "arxiv": "https://arxiv.org/abs/2610.28601",
        "code": "https://github.com/simular-ai/agent-s"
      },
      "source_hits": [
        "arxiv-harness",
        "hf-daily"
      ],
      "hf_upvotes": 23,
      "week": "2026-W44",
      "review_decision": "accept",
      "repo_id": "agent-s4",
      "review_date": "2026-10-27",
      "daily_section": "historical",
      "edition_kind": "retrospective",
      "recap_period": "2026-10-14--2026-10-27"
    },
    {
      "id": "arxiv:2610.28069",
      "arxiv_id": "2610.28069",
      "version": 1,
      "title": "Training the Whole Agent, Not Just the Model",
      "short_title": "HarnessAdapt-RL",
      "authors": [
        "Lingxiao Tang",
        "Wei Tao",
        "Yiwen Guo",
        "Lingfeng Bao"
      ],
      "published": "2026-10-20T00:00:00.000Z",
      "updated": "2026-10-20T00:00:00.000Z",
      "hf_daily_dates": [
        "2026-10-21"
      ],
      "announced": "2026-10-20",
      "announced_source": "published",
      "news_date": "2026-10-21",
      "window_status": "backlog",
      "in_repo": true,
      "gate": {
        "verdict": "harness",
        "harness_specific": "yes",
        "improved_object": "none",
        "improved_object_zh": "详见证据卡"
      },
      "direction": {
        "primary": "harness",
        "harness_sub": "coevolution",
        "secondary": []
      },
      "summary_zh": "RL 期间，一个教师生成任务相关的机制级技能，另一个根据反复出现的交互问题演化工具和 hooks；只保留使通过率或可训练率提高且另一指标不降的修改。",
      "summary_en": "During RL, one teacher generates task-related mechanism-level skills, while another evolves tools and hooks from recurring interaction problems; retains only edits that improve pass rate or trainability without lowering the other metric.",
      "evidence_quick": {
        "held_out": {
          "value": "unknown",
          "quote": null
        },
        "matched_budget": {
          "value": "unknown",
          "quote": null
        },
        "cross_model": {
          "value": "unknown",
          "quote": null
        },
        "code_released": {
          "value": "unknown",
          "quote": null
        }
      },
      "evidence_notice": "Curated evidence; see evidence card",
      "links": {
        "hf": "https://huggingface.co/papers/2610.28069",
        "arxiv": "https://arxiv.org/abs/2610.28069",
        "code": "https://github.com/lingxiaotang/HarnessAdapt-RL"
      },
      "source_hits": [
        "hf-daily"
      ],
      "hf_upvotes": 20,
      "week": "2026-W44",
      "review_decision": "accept",
      "repo_id": "harnessadapt-rl",
      "review_date": "2026-10-27",
      "daily_section": "historical",
      "edition_kind": "retrospective",
      "recap_period": "2026-10-14--2026-10-27"
    },
    {
      "id": "arxiv:2610.27435",
      "arxiv_id": "2610.27435",
      "version": 1,
      "title": "Harness Self-Evolution for Long-Horizon Terminal Agents",
      "short_title": "RiseHarness",
      "authors": [
        "Ruhan Wang",
        "Yucheng Shi",
        "Zongxia Li",
        "Junyao Yang",
        "Zhongzhi Li",
        "Yue Yu",
        "Haitao Mi",
        "Dongruo Zhou",
        "Leoweiliang"
      ],
      "published": "2026-10-20T00:00:00.000Z",
      "updated": "2026-10-20T00:00:00.000Z",
      "hf_daily_dates": [
        "2026-10-21"
      ],
      "announced": "2026-10-20",
      "announced_source": "published",
      "news_date": "2026-10-21",
      "window_status": "backlog",
      "in_repo": true,
      "gate": {
        "verdict": "harness",
        "harness_specific": "yes",
        "improved_object": "none",
        "improved_object_zh": "详见证据卡"
      },
      "direction": {
        "primary": "harness",
        "harness_sub": "evolve_main",
        "secondary": []
      },
      "summary_zh": "针对重复运行时时成时败的任务：结合 Agent 轨迹与运行时内部追踪，用 Harness Handbook 把行为映射到源码，并检索其他框架的实现作参考。",
      "summary_en": "Targets tasks with inconsistent success across repeated runs: combines agent trajectories and internal runtime traces, maps behavior to source code using a Harness Handbook, and retrieves implementations from other frameworks as references.",
      "evidence_quick": {
        "held_out": {
          "value": "yes",
          "quote": "papers/riseharness.md"
        },
        "matched_budget": {
          "value": "unknown",
          "quote": null
        },
        "cross_model": {
          "value": "unknown",
          "quote": null
        },
        "code_released": {
          "value": "unknown",
          "quote": null
        }
      },
      "evidence_notice": "Curated evidence; see evidence card",
      "links": {
        "hf": "https://huggingface.co/papers/2610.27435",
        "arxiv": "https://arxiv.org/abs/2610.27435"
      },
      "source_hits": [
        "arxiv-harness",
        "hf-daily"
      ],
      "hf_upvotes": 14,
      "week": "2026-W44",
      "review_decision": "accept",
      "repo_id": "riseharness",
      "review_date": "2026-10-27",
      "daily_section": "historical",
      "edition_kind": "retrospective",
      "recap_period": "2026-10-14--2026-10-27"
    },
    {
      "id": "arxiv:2610.27327",
      "arxiv_id": "2610.27327",
      "version": 1,
      "title": "Unsupervised Tool Evolution for Coding Agents in Data-Lake Analytics",
      "short_title": "TEAL",
      "authors": [
        "Yuqing Yang",
        "Yurong Liu",
        "Deqing Fu",
        "Manav Garg",
        "Sravan Babu Bodapati",
        "Fatma Ozcan",
        "Alon Y. Halevy",
        "Jiani Zhang"
      ],
      "published": "2026-10-19T00:00:00.000Z",
      "updated": "2026-10-19T00:00:00.000Z",
      "hf_daily_dates": [
        "2026-10-21"
      ],
      "announced": "2026-10-19",
      "announced_source": "published",
      "news_date": "2026-10-21",
      "window_status": "backlog",
      "in_repo": false,
      "gate": {
        "verdict": "harness",
        "harness_specific": "yes",
        "improved_object_zh": "工具库与用法技能"
      },
      "direction": {
        "primary": "harness",
        "harness_sub": "component",
        "secondary": []
      },
      "summary_zh": "无标签设置下跨问题演化工具，只保留 Worker 复跑时采用且有证据支持的修改",
      "summary_en": "Tools evolve across unlabeled problems; changes are retained only when the Worker adopts them on replay and the answer has supporting trajectory evidence.",
      "evidence_quick": {
        "code_released": {
          "value": "unknown",
          "quote": null
        },
        "cross_model": {
          "quote": null,
          "value": "unknown"
        },
        "held_out": {
          "quote": null,
          "value": "unknown"
        },
        "matched_budget": {
          "quote": "lifting Sonnet-5 at 15 steps by 9.3 points",
          "value": "yes"
        }
      },
      "evidence_notice": "Automated, unverified",
      "links": {
        "hf": "https://huggingface.co/papers/2610.27327",
        "arxiv": "https://arxiv.org/abs/2610.27327"
      },
      "source_hits": [
        "hf-daily"
      ],
      "hf_upvotes": 13,
      "week": "2026-W44",
      "review_decision": "accept",
      "repo_id": null,
      "review_date": "2026-10-27",
      "daily_section": "historical",
      "edition_kind": "retrospective",
      "recap_period": "2026-10-14--2026-10-27"
    },
    {
      "id": "arxiv:2610.27227",
      "arxiv_id": "2610.27227",
      "version": 1,
      "title": "Which Agent's Skill Should Change? Group-Relative Credit Assignment for Evolving Multi-Agent Teams",
      "short_title": "GRSE",
      "authors": [
        "Yanan Ma",
        "Yang Wang",
        "David Le",
        "Chenhan Yuan",
        "Yizheng Sun",
        "Guangwei Zhang",
        "Chenghao Xiao",
        "Chenghua Lin",
        "Imon Banerjee"
      ],
      "published": "2026-10-19T00:00:00.000Z",
      "updated": "2026-10-19T00:00:00.000Z",
      "hf_daily_dates": [
        "2026-10-22"
      ],
      "announced": "2026-10-19",
      "announced_source": "published",
      "news_date": "2026-10-22",
      "window_status": "backlog",
      "in_repo": false,
      "gate": {
        "verdict": "harness",
        "harness_specific": "yes",
        "improved_object_zh": "多 Agent 技能库"
      },
      "direction": {
        "primary": "harness",
        "harness_sub": "component",
        "secondary": []
      },
      "summary_zh": "不更新权重，按组内相对比较把团队奖励归因到具体技能再修订",
      "summary_en": "Relative within-group comparisons attribute team rewards to individual skills for revision without updating model weights.",
      "evidence_quick": {
        "code_released": {
          "value": "yes",
          "quote": "https://github.com/yananma1029/GRSE-Skills"
        },
        "cross_model": {
          "quote": null,
          "value": "unknown"
        },
        "held_out": {
          "quote": null,
          "value": "unknown"
        },
        "matched_budget": {
          "quote": null,
          "value": "unknown"
        }
      },
      "evidence_notice": "Automated, unverified",
      "links": {
        "hf": "https://huggingface.co/papers/2610.27227",
        "arxiv": "https://arxiv.org/abs/2610.27227",
        "code": "https://github.com/yananma1029/GRSE-Skills"
      },
      "source_hits": [
        "hf-daily"
      ],
      "hf_upvotes": 22,
      "week": "2026-W44",
      "review_decision": "accept",
      "repo_id": null,
      "review_date": "2026-10-27",
      "daily_section": "historical",
      "edition_kind": "retrospective",
      "recap_period": "2026-10-14--2026-10-27"
    },
    {
      "id": "arxiv:2610.26770",
      "arxiv_id": "2610.26770",
      "version": 2,
      "title": "HarnessPro: Recursive Harness Self-Improvement with Pareto Optimization for Performance-Cost Trade-offs",
      "short_title": "HarnessPro",
      "authors": [
        "Zhuoshi Pan",
        "ZhiYuan Feng",
        "Yukuo Cen",
        "Honglin Lin",
        "Junru Lu",
        "Zhongxia Chen",
        "Shuo Yang",
        "H. Vicky Zhao",
        "Jie Tang"
      ],
      "published": "2026-10-19T17:27:38Z",
      "updated": "2026-10-20T06:00:54Z",
      "hf_daily_dates": [
        "2026-10-20"
      ],
      "announced": "2026-10-19",
      "announced_source": "published",
      "news_date": "2026-10-20",
      "window_status": "backlog",
      "in_repo": true,
      "gate": {
        "verdict": "harness",
        "harness_specific": "yes",
        "improved_object": "none",
        "improved_object_zh": "详见证据卡"
      },
      "direction": {
        "primary": "harness",
        "harness_sub": "generalize",
        "secondary": []
      },
      "summary_zh": "无标签：用模型生成的测试评价候选，不访问标准答案或官方测试；按性能与成本的不同偏好产出 lite、medium、heavy 三档。",
      "summary_en": "Label-free: evaluates candidates with model-generated tests, without accessing standard answers or official tests; produces lite, medium, and heavy tiers for different performance–cost preferences.",
      "evidence_quick": {
        "held_out": {
          "value": "unknown",
          "quote": null
        },
        "matched_budget": {
          "value": "unknown",
          "quote": null
        },
        "cross_model": {
          "value": "unknown",
          "quote": null
        },
        "code_released": {
          "value": "unknown",
          "quote": null
        }
      },
      "evidence_notice": "Curated evidence; see evidence card",
      "links": {
        "hf": "https://huggingface.co/papers/2610.26770",
        "arxiv": "https://arxiv.org/abs/2610.26770",
        "code": "https://github.com/pzs19/HarnessPro"
      },
      "source_hits": [
        "arxiv-harness",
        "arxiv-rsi",
        "hf-daily"
      ],
      "hf_upvotes": 33,
      "week": "2026-W44",
      "review_decision": "accept",
      "repo_id": "harnesspro",
      "review_date": "2026-10-27",
      "daily_section": "historical",
      "edition_kind": "retrospective",
      "recap_period": "2026-10-14--2026-10-27"
    },
    {
      "id": "arxiv:2610.25078",
      "arxiv_id": "2610.25078",
      "version": 1,
      "title": "CoEvo-RSI: When Harnesses Evolve, Which Models Should Follow?",
      "short_title": "CoEvo-RSI",
      "authors": [
        "Jiaqi Su",
        "Guibin Zhang",
        "Jie-Jing Shao",
        "Cong Pang",
        "Tiankuo Yao",
        "Ivor Tsang",
        "Haiyan Yin",
        "Shuicheng Yan"
      ],
      "published": "2026-10-18T00:00:00.000Z",
      "updated": "2026-10-18T00:00:00.000Z",
      "hf_daily_dates": [
        "2026-10-21"
      ],
      "announced": "2026-10-18",
      "announced_source": "published",
      "news_date": "2026-10-21",
      "window_status": "backlog",
      "in_repo": true,
      "gate": {
        "verdict": "harness",
        "harness_specific": "yes",
        "improved_object": "none",
        "improved_object_zh": "详见证据卡"
      },
      "direction": {
        "primary": "harness",
        "harness_sub": "coevolution",
        "secondary": []
      },
      "summary_zh": "元 Agent 借共享演化档案双向继承历史 Harness 与模型检查点，用自身拒绝采样轨迹微调，缓解「脑体错配」。",
      "summary_en": "A meta agent uses a shared evolutionary archive for bidirectional inheritance of historical harnesses and model checkpoints, fine-tuning on its own rejection-sampled trajectories to mitigate “brain–body mismatch.”",
      "evidence_quick": {
        "held_out": {
          "value": "unknown",
          "quote": null
        },
        "matched_budget": {
          "value": "unknown",
          "quote": null
        },
        "cross_model": {
          "value": "unknown",
          "quote": null
        },
        "code_released": {
          "value": "unknown",
          "quote": null
        }
      },
      "evidence_notice": "Curated evidence; see evidence card",
      "links": {
        "hf": "https://huggingface.co/papers/2610.25078",
        "arxiv": "https://arxiv.org/abs/2610.25078",
        "code": "https://github.com/SuJQ0722/CoEvo-RSI"
      },
      "source_hits": [
        "arxiv-harness",
        "arxiv-rsi",
        "hf-daily"
      ],
      "hf_upvotes": 120,
      "week": "2026-W44",
      "review_decision": "accept",
      "repo_id": "coevo-rsi",
      "review_date": "2026-10-27",
      "daily_section": "historical",
      "edition_kind": "retrospective",
      "recap_period": "2026-10-14--2026-10-27"
    },
    {
      "id": "arxiv:2610.23479",
      "arxiv_id": "2610.23479",
      "version": 2,
      "title": "HexaShell: A Thin Harness for Exploring the Limits of VLM Agents as Robot Coding Agents",
      "short_title": "HexaShell",
      "authors": [
        "Hongcheng Gao",
        "Alex Zhou",
        "Jay Zhu",
        "Zelin Zheng",
        "Shijia Ge",
        "Xiping Li",
        "Di Wu",
        "Yazhe Wang",
        "Jianshu Zeng",
        "Xuan Shangguan",
        "Yongli Wang",
        "Yexing Wan",
        "Lingyu He",
        "Zhiqi Jia",
        "Sihang Wu",
        "Xiao He"
      ],
      "published": "2026-10-16T13:56:16Z",
      "updated": "2026-10-20T15:47:35Z",
      "hf_daily_dates": [
        "2026-10-19"
      ],
      "announced": "2026-10-16",
      "announced_source": "published",
      "news_date": "2026-10-19",
      "window_status": "backlog",
      "in_repo": false,
      "gate": {
        "verdict": "harness",
        "harness_specific": "yes",
        "improved_object_zh": "机器人感知工具代码"
      },
      "direction": {
        "primary": "harness",
        "harness_sub": "component",
        "secondary": [
          "embodied"
        ]
      },
      "summary_zh": "每次失败后读日志写或修工具，冻结后在新布局上测试",
      "summary_en": "After failures, the agent reads logs and writes or repairs perception tools; the tools are frozen before testing on new layouts.",
      "evidence_quick": {
        "code_released": {
          "value": "yes",
          "quote": "https://github.com/HexaFuture/HexaShell"
        },
        "cross_model": {
          "quote": null,
          "value": "unknown"
        },
        "held_out": {
          "quote": "test them on ten new layouts of the same seed",
          "value": "yes"
        },
        "matched_budget": {
          "quote": null,
          "value": "unknown"
        }
      },
      "evidence_notice": "Automated, unverified",
      "links": {
        "hf": "https://huggingface.co/papers/2610.23479",
        "arxiv": "https://arxiv.org/abs/2610.23479",
        "code": "https://github.com/HexaFuture/HexaShell"
      },
      "source_hits": [
        "arxiv-harness",
        "hf-daily"
      ],
      "hf_upvotes": 124,
      "week": "2026-W44",
      "review_decision": "accept",
      "repo_id": null,
      "review_date": "2026-10-27",
      "daily_section": "historical",
      "edition_kind": "retrospective",
      "recap_period": "2026-10-14--2026-10-27"
    },
    {
      "id": "arxiv:2610.22791",
      "arxiv_id": "2610.22791",
      "version": 1,
      "title": "Self-Evolving Coding Agents via Harness-Policy Optimization",
      "short_title": "Self-HPO",
      "authors": [
        "Zhepei Wei",
        "Wenlin Yao",
        "Baolin Peng",
        "Danqing Wang",
        "Qianhui Wu",
        "Hao Cheng",
        "Yu Meng",
        "Jianfeng Gao"
      ],
      "published": "2026-10-16T00:00:00.000Z",
      "updated": "2026-10-16T00:00:00.000Z",
      "hf_daily_dates": [
        "2026-10-20"
      ],
      "announced": "2026-10-16",
      "announced_source": "published",
      "news_date": "2026-10-20",
      "window_status": "backlog",
      "in_repo": true,
      "gate": {
        "verdict": "harness",
        "harness_specific": "yes",
        "improved_object": "none",
        "improved_object_zh": "详见证据卡"
      },
      "direction": {
        "primary": "harness",
        "harness_sub": "coevolution",
        "secondary": []
      },
      "summary_zh": "H-step 由当前策略分析自身失败并提出 Harness 候选；P-step 在选定 Harness 下做 RL；仍未解决的任务成为下一轮搜索集。",
      "summary_en": "In the H-step, the current policy analyzes its own failures and proposes harness candidates; in the P-step, it performs RL under the selected harness; still-unsolved tasks become the next round’s search set.",
      "evidence_quick": {
        "held_out": {
          "value": "unknown",
          "quote": null
        },
        "matched_budget": {
          "value": "unknown",
          "quote": null
        },
        "cross_model": {
          "value": "unknown",
          "quote": null
        },
        "code_released": {
          "value": "yes",
          "quote": "papers/self-hpo.md"
        }
      },
      "evidence_notice": "Curated evidence; see evidence card",
      "links": {
        "hf": "https://huggingface.co/papers/2610.22791",
        "arxiv": "https://arxiv.org/abs/2610.22791",
        "code": "https://github.com/MSR-Orchard/Self-HPO"
      },
      "source_hits": [
        "arxiv-harness",
        "arxiv-rsi",
        "hf-daily"
      ],
      "hf_upvotes": 55,
      "week": "2026-W44",
      "review_decision": "accept",
      "repo_id": "self-hpo",
      "review_date": "2026-10-27",
      "daily_section": "historical",
      "edition_kind": "retrospective",
      "recap_period": "2026-10-14--2026-10-27"
    },
    {
      "id": "arxiv:2610.22740",
      "arxiv_id": "2610.22740",
      "version": 2,
      "title": "R$^2$Orch: Recursive Self-Improvement for Agent Orchestration via Self-Play within a Skill-Internalization Loop",
      "short_title": "R²Orch",
      "authors": [
        "Mingda Zhang",
        "Yun Huang",
        "Qiang Huang",
        "Yanjin Li",
        "Rui Qian",
        "Qichao Zhao",
        "Tiesunlong Shen",
        "Na Zhao",
        "Jian Wang",
        "Xiaoying Tang"
      ],
      "published": "2026-10-15T23:21:21Z",
      "updated": "2026-10-22T15:32:16Z",
      "hf_daily_dates": [
        "2026-11-02"
      ],
      "announced": "2026-10-15",
      "announced_source": "published",
      "news_date": "2026-11-02",
      "window_status": "in_window",
      "in_repo": false,
      "gate": {},
      "direction": {
        "primary": "harness",
        "harness_sub": "coevolution",
        "secondary": [
          "model_weights"
        ]
      },
      "summary_zh": "在训练中按反馈共演化编排模型与版本化技能；部署时不携带技能。",
      "summary_en": "Co-evolves Director weights and versioned skills using execution feedback during training; deployment uses no skills.",
      "evidence_quick": {
        "held_out": {
          "value": "unknown",
          "quote": null
        },
        "matched_budget": {
          "value": "unknown",
          "quote": null
        },
        "cross_model": {
          "value": "unknown",
          "quote": null
        },
        "code_released": {
          "quote": "https://github.com/beita6969/R2Orch",
          "value": "yes"
        }
      },
      "evidence_notice": "Automated, unverified",
      "links": {
        "arxiv": "https://arxiv.org/abs/2610.22740",
        "code": "https://github.com/beita6969/R2Orch",
        "s2": "https://www.semanticscholar.org/paper/6baa95b99b084f05c311d0285aae5101800c06e6"
      },
      "source_hits": [
        "arxiv-rsi",
        "arxiv-version-check",
        "hf-daily",
        "s2-citations"
      ],
      "hf_upvotes": 31,
      "week": "2026-W45",
      "review_decision": "accept",
      "repo_id": null,
      "review_date": "2026-11-02",
      "edition_kind": "daily",
      "daily_section": "hf_today"
    },
    {
      "id": "arxiv:2610.22732",
      "arxiv_id": "2610.22732",
      "version": 1,
      "title": "TokenEconomizer: Experience-Driven Black-Box Cost Control for Coding Agents",
      "short_title": "TokenEconomizer",
      "authors": [
        "Aaron J. Li",
        "Weijia Xu",
        "Chandan Singh",
        "Zixiao Chen",
        "Jeevana Priya Inala",
        "Charlotte Siska",
        "Acar Tamersoy",
        "Bin Yu",
        "Ion Stoica",
        "Michel Galley",
        "Jianfeng Gao"
      ],
      "published": "2026-10-15T00:00:00.000Z",
      "updated": "2026-10-15T00:00:00.000Z",
      "hf_daily_dates": [
        "2026-10-19"
      ],
      "announced": "2026-10-15",
      "announced_source": "published",
      "news_date": "2026-10-19",
      "window_status": "backlog",
      "in_repo": false,
      "gate": {
        "verdict": "harness",
        "harness_specific": "yes",
        "improved_object_zh": "注入的微技能与工具集"
      },
      "direction": {
        "primary": "harness",
        "harness_sub": "component",
        "secondary": []
      },
      "summary_zh": "跨任务序列根据经验注入技能、裁剪长期无用工具，作用于 Claude Code/Codex/Copilot CLI",
      "summary_en": "Across tasks, the controller injects experience-driven micro-skills and removes persistently unused tools for Claude Code, Codex, and Copilot CLI.",
      "evidence_quick": {
        "code_released": {
          "value": "yes",
          "quote": "https://github.com/microsoft/TokenEconomizer"
        },
        "cross_model": {
          "quote": "Across Claude Code, Codex, and GitHub Copilot CLI",
          "value": "yes"
        },
        "held_out": {
          "quote": "generalize to diverse domains",
          "value": "yes"
        },
        "matched_budget": {
          "quote": null,
          "value": "unknown"
        }
      },
      "evidence_notice": "Automated, unverified",
      "links": {
        "hf": "https://huggingface.co/papers/2610.22732",
        "arxiv": "https://arxiv.org/abs/2610.22732",
        "code": "https://github.com/microsoft/TokenEconomizer"
      },
      "source_hits": [
        "arxiv-adjacent",
        "hf-daily"
      ],
      "hf_upvotes": 6,
      "week": "2026-W44",
      "review_decision": "accept",
      "repo_id": null,
      "review_date": "2026-10-27",
      "daily_section": "historical",
      "edition_kind": "retrospective",
      "recap_period": "2026-10-14--2026-10-27"
    },
    {
      "id": "arxiv:2610.18813",
      "arxiv_id": "2610.18813",
      "version": 1,
      "title": "Self-Evolving Agent Harness",
      "short_title": "Self-Evolving Agent Harness (theory)",
      "authors": [
        "Chuyang Wei",
        "Haoxiang Guan",
        "Minghang Zhu",
        "Kexin Zhou",
        "Jiyan He",
        "Xiren Zhou",
        "Huanhuan Chen",
        "Yifei Shen"
      ],
      "published": "2026-10-13T00:00:00.000Z",
      "updated": "2026-10-13T00:00:00.000Z",
      "hf_daily_dates": [
        "2026-10-16"
      ],
      "announced": "2026-10-13",
      "announced_source": "published",
      "news_date": "2026-10-16",
      "window_status": "backlog",
      "in_repo": true,
      "gate": {
        "verdict": "harness",
        "harness_specific": "yes",
        "improved_object": "none",
        "improved_object_zh": "详见证据卡"
      },
      "direction": {
        "primary": "theory_survey",
        "harness_sub": null,
        "secondary": [
          "harness"
        ]
      },
      "summary_zh": "把 Harness 自进化拆成「更新对象」与「提案/准入机制」，给出持续 PAC 学习下的泛化界、保留界与基于新鲜验证数据的准入规则。理论界依赖明确假设。",
      "summary_en": "Decomposes harness self-evolution into update targets and proposal/admission mechanisms, providing generalization bounds, retention bounds, and admission rules based on fresh validation data under continual PAC learning. The theoretical bounds depend on explicit assumptions.",
      "evidence_quick": {
        "held_out": {
          "value": "unknown",
          "quote": null
        },
        "matched_budget": {
          "value": "unknown",
          "quote": null
        },
        "cross_model": {
          "value": "unknown",
          "quote": null
        },
        "code_released": {
          "value": "unknown",
          "quote": null
        }
      },
      "evidence_notice": "Curated evidence; see evidence card",
      "links": {
        "hf": "https://huggingface.co/papers/2610.18813",
        "arxiv": "https://arxiv.org/abs/2610.18813",
        "code": "https://github.com/zgcagi/Harness-RSI"
      },
      "source_hits": [
        "hf-daily"
      ],
      "hf_upvotes": 101,
      "week": "2026-W44",
      "review_decision": "accept",
      "repo_id": "self-evolving-agent-harness",
      "review_date": "2026-10-27",
      "daily_section": "historical",
      "edition_kind": "retrospective",
      "recap_period": "2026-10-14--2026-10-27"
    },
    {
      "id": "arxiv:2610.17860",
      "arxiv_id": "2610.17860",
      "version": 1,
      "title": "Socratic Skill Learning: Interleaving Teaching and Practice for Transferable Agent Skills",
      "short_title": "Socratic Skill Learning",
      "authors": [
        "Letian Zhu",
        "Han Li",
        "Yifan Yao",
        "Xinping Lei",
        "Xinyu Che",
        "Jiahao Wang",
        "Dianyi Wang",
        "Yucheng Fu",
        "Jiakai Wang",
        "Zhiqi Bai",
        "Wenbo Su",
        "Jiaheng Liu"
      ],
      "published": "2026-10-13T00:00:00.000Z",
      "updated": "2026-10-13T00:00:00.000Z",
      "hf_daily_dates": [
        "2026-10-22"
      ],
      "announced": "2026-10-13",
      "announced_source": "published",
      "news_date": "2026-10-22",
      "window_status": "backlog",
      "in_repo": false,
      "gate": {
        "verdict": "harness",
        "harness_specific": "yes",
        "improved_object_zh": "可迁移的 Skill 文件"
      },
      "direction": {
        "primary": "harness",
        "harness_sub": "component",
        "secondary": []
      },
      "summary_zh": "Student 通过实践编写并修订 Skill，在新会话的留出用例上复用",
      "summary_en": "The student writes and revises transferable Skill files through practice, then reuses them on held-out cases in new sessions.",
      "evidence_quick": {
        "code_released": {
          "value": "yes",
          "quote": "https://github.com/NJU-LINK/Socratic-Skill-Learning"
        },
        "cross_model": {
          "quote": "Across three model pairs on shared families",
          "value": "yes"
        },
        "held_out": {
          "quote": "held-out evaluation in fresh solver sessions",
          "value": "yes"
        },
        "matched_budget": {
          "quote": null,
          "value": "unknown"
        }
      },
      "evidence_notice": "Automated, unverified",
      "links": {
        "hf": "https://huggingface.co/papers/2610.17860",
        "arxiv": "https://arxiv.org/abs/2610.17860",
        "code": "https://github.com/NJU-LINK/Socratic-Skill-Learning"
      },
      "source_hits": [
        "hf-daily"
      ],
      "hf_upvotes": 49,
      "week": "2026-W44",
      "review_decision": "accept",
      "repo_id": null,
      "review_date": "2026-10-27",
      "daily_section": "historical",
      "edition_kind": "retrospective",
      "recap_period": "2026-10-14--2026-10-27"
    },
    {
      "id": "arxiv:2610.15915",
      "arxiv_id": "2610.15915",
      "version": 1,
      "title": "VIEWCRAFT: Evolving What Multimodal Agents See and How They Read It",
      "short_title": "ViewCraft",
      "authors": [
        "Binxiao Xu",
        "Ruichuan An",
        "Bocheng Zou",
        "Weihua Du",
        "Xinyu Wei",
        "Hangyu Guo",
        "YiFan Zhang",
        "Renye Yan",
        "Chengqi Duan",
        "Hang Hua"
      ],
      "published": "2026-10-11T00:00:00.000Z",
      "updated": "2026-10-11T00:00:00.000Z",
      "hf_daily_dates": [
        "2026-10-14"
      ],
      "announced": "2026-10-11",
      "announced_source": "published",
      "news_date": "2026-10-14",
      "window_status": "backlog",
      "in_repo": false,
      "gate": {
        "verdict": "harness",
        "harness_specific": "yes",
        "improved_object_zh": "证据接口与读取技能"
      },
      "direction": {
        "primary": "harness",
        "harness_sub": "component",
        "secondary": []
      },
      "summary_zh": "失败轨迹驱动修订，回放与配对评估决定保留，同 token 预算下仍有优势",
      "summary_en": "Failure traces drive revisions to evidence interfaces and reading skills; replay and paired evaluation select retained changes, with gains under a matched token budget.",
      "evidence_quick": {
        "code_released": {
          "value": "yes",
          "quote": "https://github.com/Little-Fridge/ViewCraft"
        },
        "cross_model": {
          "quote": "The learned skills also transfer across executors",
          "value": "yes"
        },
        "held_out": {
          "quote": null,
          "value": "unknown"
        },
        "matched_budget": {
          "quote": "The advantage persists at a matched input-token budget.",
          "value": "yes"
        }
      },
      "evidence_notice": "Automated, unverified",
      "links": {
        "hf": "https://huggingface.co/papers/2610.15915",
        "arxiv": "https://arxiv.org/abs/2610.15915",
        "code": "https://github.com/Little-Fridge/ViewCraft"
      },
      "source_hits": [
        "hf-daily"
      ],
      "hf_upvotes": 24,
      "week": "2026-W44",
      "review_decision": "accept",
      "repo_id": null,
      "review_date": "2026-10-27",
      "daily_section": "historical",
      "edition_kind": "retrospective",
      "recap_period": "2026-10-14--2026-10-27"
    },
    {
      "id": "arxiv:2610.14224",
      "arxiv_id": "2610.14224",
      "version": 1,
      "title": "Harness^2: Recursive Agent Harnessing for an Open World",
      "short_title": "Harness²",
      "authors": [
        "Ruiyao Xu",
        "Yanfei Chen",
        "Zhongying CuiZhu",
        "Bhavana Dalvi Mishra",
        "Yifei Ming",
        "Han Yu",
        "Rujun Han",
        "Chen-Yu Lee",
        "Tomas Pfister"
      ],
      "published": "2026-10-09T00:00:00.000Z",
      "updated": "2026-10-09T00:00:00.000Z",
      "hf_daily_dates": [
        "2026-10-14"
      ],
      "announced": "2026-10-09",
      "announced_source": "published",
      "news_date": "2026-10-14",
      "window_status": "backlog",
      "in_repo": true,
      "gate": {
        "verdict": "harness",
        "harness_specific": "yes",
        "improved_object": "none",
        "improved_object_zh": "详见证据卡"
      },
      "direction": {
        "primary": "harness",
        "harness_sub": "improver",
        "secondary": []
      },
      "summary_zh": "开放世界、无真值验证器：任务层用「提出—试探—组合」对比步引导编辑，上层把编辑经验沉淀到持续改进 Harness。",
      "summary_en": "Open world without a ground-truth verifier: the task level guides edits through “propose–probe–compose” contrastive steps; the upper level distills editing experience into a continually improving harness.",
      "evidence_quick": {
        "held_out": {
          "value": "unknown",
          "quote": null
        },
        "matched_budget": {
          "value": "unknown",
          "quote": null
        },
        "cross_model": {
          "value": "unknown",
          "quote": null
        },
        "code_released": {
          "value": "unknown",
          "quote": null
        }
      },
      "evidence_notice": "Curated evidence; see evidence card",
      "links": {
        "hf": "https://huggingface.co/papers/2610.14224",
        "arxiv": "https://arxiv.org/abs/2610.14224",
        "code": "https://github.com/google-research/harness2"
      },
      "source_hits": [
        "hf-daily"
      ],
      "hf_upvotes": 124,
      "week": "2026-W44",
      "review_decision": "accept",
      "repo_id": "harness2",
      "review_date": "2026-10-27",
      "daily_section": "historical",
      "edition_kind": "retrospective",
      "recap_period": "2026-10-14--2026-10-27"
    }
  ]
}