llm-application · 知识库活文档
- 更新：v124 8 件新 repo_card 入正文(6 外壳主栖 + 2 partial)+ 0 arXiv/paper_card 净增 + 8 repo URL 入池;新增"外壳层接入点轴"
0. 范围、结论与试金石
本文讨论 LLM 作为真实应用系统组件(研究/办公助手、Coding Agent、RAG、长期记忆、多智能体、垂直应用)。试金石(全篇判定基准,不可变更):任务真实可复现 · 证据链可审计 · 跨会话状态可治理 · 失败可分解定位 · 成本与安全满足生产约束。
v124 本轮结论一句话:工作集 8 件 repo_card 中 6 件不涉及模型与训练,只解决"把模型接进真实工作流"的最后一公里 —— Agent 应用外壳(shell)层正在独立成一等工程品类,与 v116-v123 已立的"训练侧 harness"形成镜像。
v117 骨架基线(承重,详见 changelog v117 段):net-new 0 件主分类 arXiv + 18 件承接级候选合入 v116 已立骨架(HarnessDev / AgentJudgeBench / EvoGenUI-Bench / Aspire / Harness-of-Harness / Hebero-vs-RobotWorld 范式二分 / AgentSysBench / Mem0-MemOS-Cognee-MemSearch 四足鼎立 / REMORY + galahad-kv / Memento 3 / Skill Constellations / State-Bench / HarnessSQL / RAG 文档结构消融 + HNSW 几何 + Q-RAG + 四轴分类 / Kiln on Trainium / Mem0 LoCoMo / Meena 2001.09977 立标池历史回填第 1 例)+ GLM-5.3 上线 Amazon Bedrock 双源确认(753B 总参 / 40B active · MoE · 1M-token context · cross-Region inference profiles · reasoning effort 可调)+ HF Daily MiMo-V2.6 登顶(国内厂商首次)+ frontier lab 商业模式三栖叙事 + Microsoft Agent 框架生态成熟化拐点双锚定。v122 基线:net-new 3 件 arXiv(OmniCapBench + Inherit-MAS + Self-Retrospection Distillation)+ 8 paper_card。v121 基线:net-new 1 件 arXiv(Meena)+ 7 repo_card。
1. 现状全景
1.1 立标池与承接稳态
HF Daily 立标池结构回稳期延续(MiMo-V2.6 登顶为国内厂商首次)+ 评测栖预备 v116 37 → v117 41 → v122 42 扩位级(HarnessDev / AgentJudgeBench / EvoGenUI-Bench / Aspire 依次加入)+ RAG Defense 轴 v116 7 → v117 11 件预备扩增稳态 + Agent 安全四栖(行为/推理/输出/协议)预备稳态 + Memory 第十六栖"分化基础设施栈"预备邻接第 1 例。
1.2 Harness 簇与训练侧骨架
Harness 设计自动化四路径 ⚠⚬ = HarnessDev(LLM 生成 harness)+ EMHO(轨迹驱动 harness 演进)+ Mara Chain(被拒候选重用)+ Harness-of-Harness(harness 的 harness 持续改进)。加上 HarnessSQL(原生 SQL Agent 训练)、Agent Lightning v1.0(Qwen3.5-9B SWE-bench Verified 41.8% → 56.4%)、MXC 政策驱动隔离。训练侧 harness 已成体系。
1.3 评测范式
Hebero vs RobotWorld 评测范式二分 = "train 侧 vs eval 侧"形式化 ⚠⚬⚬⚬;RobotWorld 84 任务 63 未解 + RoboQuest 23.2% + SWE-Game 247 任务 <60/100 = frontier lab 模型在长程/具身任务仍有大量失败空间。
1.4 【v124 新增】外壳层"接入点轴" ⚠⚬
v124 工作集 8 件中 6 件属于外壳而非模型能力,按接入点归为六类,每类同棒位出现独立开源实现:
- ① 检索接入 — xerj(单 Rust 二进制索引 code/docs/logs/JSONL/CSV/PDF;BM25 + 向量 + 混合;免 grep 代码搜索;agent memory 与 RAG;Elasticsearch 兼容 API;Corpus Hub 数百参考记忆语料)
- ② 对话/网页接入 — synaplan(PHP/Symfony 自托管;任意模型聊天;自有文档 RAG;可嵌入 widget;MCP;逐条回答成本可见)+ geo-doctor(装进 AI 助手的 GEO 诊断 skill)
- ③ 文档协作接入 — decks(幻灯片下留言 → Claude Code / Codex 改 → 在该幻灯片回帖)+ Thesis-AI(给定知识库 → 规范学术散文)
- ④ 语音/电信接入 — Lampline(导入 Dialogflow CX / CX Agent Studio agent + 自有硬件经 jambonz 接听电话 + 20 家模型供应商)
- ⑤ 学术检索接入 — semantic-scholar-mcp(14 个类型化工具:论文检索/引用图谱/作者画像/推荐;PyPI + Docker)
- ⑥ 记忆基础设施接入 — MemT-1.0(官方仓库)
这一层在 v116-v123 骨架中此前未被显式命名。已有骨架按能力轴组织(Memory 多栖 / RAG 多栖 / Agent 安全四栖 / 评测多栖),本轮提示一条正交轴:按接入点分类。"接进哪个界面"正成为与"用什么模型"同等重要的工程决策 ⚠⚬。
方法学警示 ⚠⚬:本轮 8 件全部周增 +0,其中 5 件为 2026-10-20 首次采集。周增全零且多为首次采集 = GitHub Trending 采样窗口的噪声日信号,不能据此判断热度衰减;唯一有真实增长的是 xerj(+12131/周)。
1.5 语言栈去 Python 化第 3 例
v124 语言分布 Rust 1 / PHP 1 / HTML 1 / JavaScript 1 / Python 3 / TypeScript 1。PHP/Symfony 承载完整 agent 平台是首次出现的非 Python/TS 技术栈,与 v121-v123 的 Bun + SQLite、单 Rust 二进制共同构成去 Python 化三例证据 ⚠⚬。
2. 共识
- Agent 框架生态成熟化拐点双锚定确立 ⚠⚬⚬⚬:Agent Framework 1.0 GA + Agent Lightning v1.0 + MXC + AutoGen 维护模式 + LangChain/LangGraph 全栈统一。
- frontier lab 商业模式三栖叙事 ⚠⚬⚬⚠:Anthropic 安全/治理产品化 + OpenAI 成本/效率优化 + DeepMind 多模态/隐私 = 从模型 API 走向"模型 + 安全 + 治理"三维订阅。
- Harness 设计自动化四路径 ⚠⚬⚬。
- MiMo-V2.6 登顶 HF Daily 立标池 ⚠⚬⚬:国内厂商首次。
- 【v124 新增】外壳层三命题 ⚠⚬:
- 命题 A 本地优先是外壳层默认架构选择 — xerj 单 Rust 二进制 + Elasticsearch 兼容;quireink 一个 Bun 进程两个 SQLite 无云端账号;synaplan 自托管 + 可嵌入 widget。三团队独立收敛到"自托管 + 本地索引 + 标准 API 兼容"。这与 rogue agent / HF breach / MCP 无认证等安全事件形成因果闭环:数据不出本地同时消解最大一类 Agent 安全面。
- 命题 B 外壳层第一 KPI 是成本可见性而非能力 — synaplan 明写"see the cost of every answer";xerj 40× token 节省;geo-doctor 是成本归因工具。与 neatlogs/neatroots 可观测性栖 + AgentSysBench"模型推理不再是主要成本中心"方向一致。
- 命题 C 外壳层正在绕过 MCP 生态另起一套 — xerj 走 Elasticsearch 兼容 API;synaplan 走 widget + MCP 双轨;semantic-scholar-mcp 走纯 MCP;geo-doctor 走"skill 装进助手"。MCP 作为协议已确立,但作为外壳打包分发方式尚未确立 ⚠⚬。
3. 争议
#138(预备级):沿用 v117 §0 七项锚定。v124 未新增可独立成项的争议,但"接入点轴"与既有能力轴存在重叠(xerj 同属 RAG 栖与检索接入;semantic-scholar-mcp 同属 academic discovery 栖与学术检索接入)。处理方式:能力轴为主干(§6 按栖编号),接入点轴作为横切视图(§1.4 + §5),同一材料在两处出现但只计入一次净增 ⚠⚬。
4. 开放问题
承重 10 项(Q117.560-Q117.569):① HarnessDev 四路径相对适用边界(代码 vs 写作 vs ML 实验)与 HarnessSQL 承接关系;② AgentJudgeBench DAG judge 失效与 MIST 的机制差异 + DAG 深宽度分级;③ EvoGenUI-Bench 自动 vs 人工评分体系 + 任务分布;④ Aspire 模糊目标与 Agent Plasticity 曲线互用评测边界;⑤ Harness-of-Harness 多日 70+ 迭代可复现性;⑥ Hebero vs RobotWorld 二分与 "1 episode / 模型-任务" 重新解读边界;⑦ AgentSysBench 成本中心结论的工业普适性 + disaggregated serving -29-40% 场景数字;⑧ REMORY + galahad-kv 方法学双轨工程边界;⑩ frontier lab 治理升档过快风险 + MiMo-V2.6 立标池判定方法学。
【v124 新增 Q124.571-Q124.577】: - ⑪ 外壳层接入点轴是否会形成独立稳定分类:六类接入点单日同时出现独立实现,是 Trending 采样偏差还是真实一等品类信号?需连续 3-5 棒观察同类外壳件是否持续出现。 - ⑫ xerj 单周 +12131 stars 的成因归因:MCP 搜索协议标准化的真实需求爆发,还是单点传播?其 Corpus Hub 是否构成 Agent 记忆的可复用分发层,进而与 Memory 各栖竞争同一生态位? - ⑬ BM25 + 向量 + 混合 + 代码搜索 四合一收益是否被夸大:v120 记录的 40× token 节省是相对 grep 的对比,而非相对向量数据库的对比,两者成本口径不同,不可直接相减 ⚠⚬。 - ⑭ 外壳层与 MCP 协议的竞合边界:四种分发形态(API 兼容 / MCP / widget / skill)长期收敛到哪一种? - ⑮ Geo-Doctor 这类 GEO skill 是否具备可核查判定标准:仓库卡只给流程(查清为什么不推荐 → 改资料 → 写内容 → 再复查),未给任何指标定义;GEO"推荐率"与 §0 试金石"证据链可审计"如何对齐 ⚠⚬。 - ⑯ Lampline 迁移 Dialogflow CX 的逆向兼容边界:导入 / 自有硬件接听 / 20 供应商三条同时成立时,超时/打断/转接/语音特性如何验证?语料卡未提供对照评测。 - ⑰ 学术工具池是否已到可评估规模:v119-v123 已积累 Imbad0202 / aipoch / SakanaAI / KalarisLabs / Aitherium 五件,加本轮两件共七件,但 star 数 2 至 85 不等、周增全零,尚无可横向比较的效果学证据,不应升档为立标。
5. 趋势判断
① frontier lab 商业生态结构性反转 + 开源主权模型 enterprise tier 双锚定(GLM-5.3 Bedrock)⚠⚬⚬;② Agent 框架生态成熟化拐点沿用稳态 ⚠⚬⚬⚬;③ ME-World 多 agent 同步 ego-stream 立标候选 + Hebero vs RobotWorld 二分形式化;④ 评测栖 v117 41 → v122 42 扩位级 + RAG Defense 轴 + Agent 安全四栖 + Memory 第十六栖锚点;⑤ AI 工程前沿基线预备扩增稳态 + frontier lab 公告密度连续回升 + P0 警示延续;⑥ agent 自我改进 + 记忆压缩三学派(Memento 3 语义外部化 + REMORY 神经压缩 + MiMo-V2.6 训练自我改进)⚠⚬⚬⚠;⑦ Tom 10-12 雷达 13 候选归位;⑧ 【v124】Agent 应用外壳层独立成一等工程品类 ⚠⚬:六类接入点各有独立实现 + 本地优先默认化 + 成本可见性第一 KPI + 分发形态未收敛;⑨ 【v124】外壳层与 Memory 生态位开始重叠 ⚠⚬:xerj 宣称 "agent memory and RAG" 且 Corpus Hub 提供数百参考记忆语料,与 Memory 第十六栖(Mem0/MemOS/Cognee/MemSearch)、第十七栖(通信开销型压缩)、第十八栖(具身文献发现)存在生态位竞争,这是本轮最值得后续 3-5 棒跟踪的张力 ⚠⚬⚬。
6. 工程落地框架
① RAG Defense 七栖 → 十一栖预备扩增稳态(含 RAG 文档结构消融 arXiv:2610.10170 + HNSW 几何理论 arXiv:2610.12312 + Q-RAG arXiv:2511.07328 ICLR 2026 + RAG 四轴分类法 arXiv:2610.01936);② Agent 安全四栖(行为/推理/输出/协议);③ Agent 可靠性栖(UndoBench + SWE-Serve + CUAWright + LongHarness Bench + ThinkingBox);④ 长上下文栖(Behavior-Preserving KV Cache + HLA + The Extender + UNREAL + 50M Token Window);⑤ Memory 十二 → 十六栖(Memadapter + Selection-vs-Extraction + In-Parameter Memory + DyadMem URAM + MemFold + Honeycomb + ExperienceIndex + SkillForge + Real Long-Term Memory + REMORY 第十五栖 + Mem0/MemOS/Cognee/MemSearch 第十六栖 + LoCoMo 91.6%/72.9%/10×/~1.44s);⑥ reasoning 主轴(Five Requirements + R1-R5);⑦ Coding Agent 工具半衰期栖(Code Detector arXiv:2610.01664 ASE 2026 + FastOPD VLA);⑧ RAG 范式跃迁 REC ↔ COMP(RECAST arXiv:2610.10507 + Agentic AutoRAG + EC-RAG + UNREAL);⑨ 知识工程栖(awesome-llm-knowledge-systems:RAG + Context Eng + Harness Eng + Skill Systems + Agent Memory + MCP + Progressive Disclosure);⑩ Agent 框架生态成熟化拐点双锚定;⑪ ME-World KAIST 多 agent 同步 ego-stream;⑫ GLM-5.3 Bedrock enterprise tier 双锚定;⑬ jay engineering 5 联 AI 工程前沿基线;⑭ Agent 评测新维度 11 件(Memento 3 + Agent Plasticity + Opera + SafeActBench + EMHO + AgencyBench v2 + LongHarness Bench + HarnessDev + AgentJudgeBench + EvoGenUI-Bench + Aspire);⑮ Skill Constellations arXiv:2610.11169;⑯ Opera arXiv:2609.33987 Verbal Critic 长程 Coding Agent;⑰ Hebero vs RobotWorld 二分形式化;⑱ AgentSysBench arXiv:2608.15127 178,799 session 成本模型重定义(推理不再是主要成本中心 · sandbox/搜索/embedding/向量库操作占主导 · disaggregated serving -29-40%);⑲ Harness 四路径;⑳ State-Bench 450 企业任务 + HarnessSQL;㉑ RAG 工程层/理论层双立标;㉒ MiMo-V2.6 + 三学派记忆压缩;㉓ Tom 10-12 雷达 13 候选(AAArena + SoftwareGym + Work2Skill + Orthros + Verifiable RAG + Opera + ORCAGen + Skill Constellations + Mara Chain + Hebero + Co-Evolved Communication + Behavioral Persistence + δ-mem)。
㉔ 6 件 net-new repo_card(v120) ⚠⚬⚬⚬ = VideoJev(multimodal 栖预备)+ xerj(RAG 栖 + Memory 第十七栖"通信开销型 Memory 压缩")+ mcp-accessibility-scanner(MCP 协议层 skill 第 2 例)+ neatlogs/neatroots(Agent 可观测性栖第 1 例)+ PaperBot(academic-writing acceptance)+ arxiv-digest(RAG + Tools)。
【v124 新增】㉕-㉜ · 外壳层"接入点轴"工程化锚入(8 件 · 2026-10-20) ⚠⚬ — 与 §1-§5 能力轴为横切关系,同料只计一次净增。本轮 0 arXiv / 0 paper_card / 0 net-new CVE / 0 net-new DOI。
- ㉕ xerj-org/xerj(card
organized/repo_cards/4393-xerj-org-xerj.md· https://github.com/xerj-org/xerj · 10669★ +12131/周 Apache-2.0 Rust · production · 首次采集 10-15)= v120 主轴 anchor 沿用稳态第 3 例,且本轮发生实质范围扩张 ⚠⚬⚬⚠。对比 v120(5902★ +3549,"XERJ 数据搜索新范式")与 v123(8618★ +2401,第 2 例),本轮 card 首次给出完整能力矩阵。三处实质变化:① 形态从"搜索引擎"扩为"搜索引擎 + MCP server",进入命题 C 的四路分发之争;② 语料从 code/docs 扩至 logs/JSONL/CSV/PDF,BM25 的结构化半文本优势域显著扩大;③ Corpus Hub 是 Memory 生态位新竞争者(§5⑨)。单周 +12131 ≈ v123 结束时存量 8618 的 1.4 倍,是本轮唯一真实增长信号,也是 v120 "40× token 节省"承诺最完整的一次溯源。 - ㉖ metadist/synaplan(card
4504-metadist-synaplan.md· https://github.com/metadist/synaplan · 201★ +0 Apache-2.0 PHP · experimental · 首采 10-20)= 自托管 AI 平台"对话/网页接入"栖预备第 1 例 ⚠⚬。PHP/Symfony 栈;任意模型聊天;自有文档 RAG;可嵌入 widget;MCP;逐条回答成本可见;德国出品。技术栈选择构成本轮去 Python 化证据之一(§1.5)。 - ㉗ paulmil11/decks(card
4503-paulmil11-decks.md· https://github.com/paulmil11/decks · 85★ +0 MIT HTML · experimental · 首采 10-20)= Agent 优先文档协作接入点栖预备第 1 例 ⚠⚬,且是第 3 件"在既有工件上留言、由 coding agent 回改"的交互范式件。范式:幻灯片下留言 → Claude Code / Codex 直接改 → 在该幻灯片上回复。与 v123 awesome-short-drama-skills 并置,说明"Agent 作为协作者嵌入人类既有工作面(而非独立对话框)"正成为独立交互品类 ⚠⚬。HTML 为主语言表明本体是静态前端 + 外部 agent CLI 驱动,shell 极薄、agent 在宿主侧。 - ㉘ Fangx-AI/geo-doctor(card
4502-fangx-ai-geo-doctor.md· https://github.com/Fangx-AI/geo-doctor · 90★ +0 MIT JavaScript · experimental · 待LLM分类=是 · 首采 10-20)= GEO"增长/归因接入点"栖预备第 1 例 ⚠⚬,中文开源、中英双语 GEO skill。四步工作流:一句话装进 AI 助手 → 查清豆包/DeepSeek/ChatGPT/Claude 为什么不推荐你 → 改资料、写内容 → 再复查。唯一带"待 LLM 分类"标记入池,且无任何量化指标(§4 Q124.575),按试金石"证据链可审计"作 partial anchor,不升档 ⚠⚬。概念上把 §6⑱ 的成本结论翻转到收入侧:不只是"推理不再是主要成本中心",而是"被推荐与否本身成为可优化对象"。 - ㉙ C4rohan/Lampline(card
4498-c4rohan-lampline.md· https://github.com/C4rohan/Lampline · 118★ +0 Apache-2.0 Python · experimental · 首采 10-20)= 自托管语音 Agent 平台"语音/电信接入点"栖预备第 1 例 ⚠⚬,同时是本轮最重的存量系统迁移样本。四条能力:导入既有 Dialogflow CX 与 CX Agent Studio agent;构建与测试;在自有硬件经 jambonz 接听电话;20 家模型供应商。后两条关键:"自有硬件接听"把语音 Agent 从 SaaS 拉回 on-prem(与命题 A 一致),"20 家模型供应商"是模型无关性主张的第一件外壳材料。迁移保真度未决(§4 Q124.576)。 - ㉚ MemTensor/MemT-1.0(card
4497-memtensor-memt-1-0.md· https://github.com/MemTensor/MemT-1.0 · 98★ +0 Python · 许可 NOASSERTION · experimental · 待LLM分类=是 · 首采 10-20)= partial anchor ⚠⚬。card 简介仅一句 "Official repository for MemT-1.0",无技术描述、无参数、无评测。归属具信号价值:MemTensor 是已知 mem 层系统团队(org/Mem-1.6/MemOS生态),故这是已进入 Knowledge Base 的 Memory 主线官方仓库而非第三方实验。在 §5⑨ 的生态位竞争中,它与 xerj Corpus Hub 相对:自建记忆栈 vs 可复用记忆语料分发。因信息量不足以技术定位,仅作存在性锚点 ⚠⚬。 - ㉛ smaniches/semantic-scholar-mcp(card
1386-smaniches-semantic-scholar-mcp.md· https://github.com/smaniches/semantic-scholar-mcp · 23★ +0 MIT Python · experimental · 首采 08-11 → 学术复核 09-06)= academic discovery 栖 v124 第三例 anchor ⚠⚬,把 v121 双 anchor(arxiv-compass + Arxiv-daily)与 v123 embodied-daily 之外补出第四型:类型化 API 封装(前三型为 RSS 聚合 / 主题跟踪 / 具身专项)。card 带完整学术元数据(主任务 literature-search · 阶段 discovery · 状态 accepted · 方向 search · 用途 literature-discovery · 能力分面 citation/mcp/writing · 相关度 95 · 判定 deterministic · 证据 8 项),本轮唯一具备完整学术复核档案者。"14 个类型化工具"与命题 C 相关:MCP 工具是否应强类型,直接影响失败可分解定位(试金石第四项)。 - ㉜ 0xward/Thesis-AI(card
1346-0xward-thesis-ai.md· https://github.com/0xward/Thesis-AI · 2★ +0 MIT TypeScript · experimental · 首采 08-11 → 学术复核 10-20)= academic-writing 栖 v124 第四例 anchor,规模不足不升档 ⚠⚬(§4 Q124.577)。学术元数据完整:主任务 paper-drafting · 阶段 writing · 状态 accepted · 方向 write · 用途 writing-typesetting · 能力分面 writing · 相关度 80 · 判定 deterministic。技术定位:把用户提供的知识库合成为结构化、格式规范的学术散文,自述"高性能自主研究 Agent"。关键区分:它做"知识库 → 散文"单向合成,输入由用户给定,而非像 aipoch/open-science / SakanaAI/AI-Scientist-v3 那样自主检索取证。这一"受控输入 vs 自主取证"的分野,正是试金石"证据链可审计"的分界线 ⚠⚬。
6.1 v124 汇总
| 材料 | 接入点 | 能力轴归入 | 规模信号 | 处理 |
|---|---|---|---|---|
| xerj-org/xerj | 检索 | RAG 栖 / Memory 第十七栖 | 10669★ +12131 | 主轴沿用第 3 例 ⚠⚬⚬⚬ |
| metadist/synaplan | 对话/网页 | Agent 平台栖 | 201★ +0 | 栖预备第 1 例 ⚠⚬ |
| paulmil11/decks | 文档协作 | Agent 协作范式 | 85★ +0 | 栖预备第 1 例 ⚠⚬ |
| Fangx-AI/geo-doctor | 增长归因 | RAG + academic | 90★ +0 待分类 | partial ⚠⚬ |
| C4rohan/Lampline | 语音/电信 | Agent 平台栖 | 118★ +0 | 栖预备第 1 例 ⚠⚬ |
| MemTensor/MemT-1.0 | 记忆基础设施 | Memory 主线 | 98★ +0 待分类 | partial(信息量不足)⚠⚬ |
| smaniches/semantic-scholar-mcp | 学术检索 | academic discovery | 23★ +0 学术档案完整 | 第四型 anchor ⚠⚬ |
| 0xward/Thesis-AI | 文档协作 | academic-writing | 2★ +0 学术档案完整 | 不升档 ⚠⚬ |
7. 引用完整性附录### 7.1 arXiv ID 完整集合(v122 净增 3 件 · 1334 件总览 · 含 v121 沿用 1331 件 + v122 net-new 3 件)
aarXiv:2304.07193 arXiv:2310.11703 arXiv:2311.08596 arXiv:2401.16745 arXiv:2403.07652 arXiv:2406.02818 arXiv:2409.10102 arXiv:2412.15529 arXiv:2501.05444 arXiv:2501.09136 arXiv:2501.17399 arXiv:2501.19139 arXiv:2502.02276 arXiv:2502.05167 arXiv:2502.05171 arXiv:2502.07115 arXiv:2502.08826 arXiv:2503.06728 arXiv:2503.09516 arXiv:2503.12646 arXiv:2503.22911 arXiv:2503.29231 arXiv:2504.09554 arXiv:2504.11320 arXiv:2504.12330 arXiv:2504.14693 arXiv:2504.14891 arXiv:2504.19874 arXiv:2505.07833 arXiv:2505.17086 arXiv:2505.19481 arXiv:2505.22571 arXiv:2505.23723 arXiv:2505.24238 arXiv:2506.06252 arXiv:2506.06266 arXiv:2506.09498 arXiv:2506.19544 arXiv:2506.20869 arXiv:2507.00379 arXiv:2507.06608 arXiv:2507.28074 arXiv:2507.28802 arXiv:2508.02611 arXiv:2508.06600 arXiv:2508.08438 arXiv:2508.09442 arXiv:2508.09736 arXiv:2509.23040 arXiv:2510.04618 arXiv:2510.05381 arXiv:2510.08544 arXiv:2510.09665 arXiv:2510.13910 arXiv:2510.15253 arXiv:2511.01545 arXiv:2511.01815 arXiv:2511.04570 arXiv:2511.07663 arXiv:2511.16681 arXiv:2511.17729 arXiv:2512.04388 arXiv:2512.05411 arXiv:2512.09695 arXiv:2512.12167 arXiv:2512.13564 arXiv:2512.14629 arXiv:2001.09977 arXiv:2601.00227 arXiv:2601.00273 arXiv:2601.00536 arXiv:2601.03236 arXiv:2601.04043 arXiv:2601.06037 arXiv:2601.06112 arXiv:2601.06352 arXiv:2601.06362 arXiv:2601.07504 arXiv:2601.07711 arXiv:2601.08225 arXiv:2601.18137 arXiv:2601.19139 arXiv:2601.19827 arXiv:2601.21557 arXiv:2601.23254 arXiv:2602.00288 arXiv:2602.02007 arXiv:2602.02276 arXiv:2602.03442 arXiv:2602.05014 arXiv:2602.05152 arXiv:2602.06052 arXiv:2602.07962 arXiv:2602.09725 arXiv:2602.10133 arXiv:2602.10238 arXiv:2602.11224 arXiv:2602.11443 arXiv:2602.11671 arXiv:2602.13647 arXiv:2602.14374 arXiv:2602.15763 arXiv:2602.16666 arXiv:2602.16763 arXiv:2602.16901 arXiv:2602.18998 arXiv:2602.21548 arXiv:2602.21566 arXiv:2602.23368 arXiv:2603.00873 arXiv:2603.03251 arXiv:2603.03780 arXiv:2603.03781 arXiv:2603.04428 arXiv:2603.04445 arXiv:2603.04474 arXiv:2603.05451 arXiv:2603.06503 arXiv:2603.06621 arXiv:2603.06728 arXiv:2603.07379 arXiv:2603.07670 arXiv:2603.10726 arXiv:2603.10765 arXiv:2603.12646 arXiv:2603.12707 arXiv:2603.13358 arXiv:2603.15569 arXiv:2603.16104 arXiv:2603.18272 arXiv:2603.18892 arXiv:2603.20397 arXiv:2603.20432 arXiv:2603.20847 arXiv:2603.21354 arXiv:2603.21489 arXiv:2603.21972 arXiv:2603.22911 arXiv:2603.23448 arXiv:2603.25152 arXiv:2603.26498 arXiv:2603.26670 arXiv:2603.27918 arXiv:2603.28052 arXiv:2603.28583 arXiv:2603.29010 arXiv:2603.29231 arXiv:2604.00901 arXiv:2604.01395 arXiv:2604.01647 arXiv:2604.01707 arXiv:2604.06185 arXiv:2604.08571 arXiv:2604.09666 arXiv:2604.11978 arXiv:2604.14222 arXiv:2604.16548 arXiv:2604.17227 arXiv:2604.18234 arXiv:2604.19157 arXiv:2604.21413 arXiv:2604.22748 arXiv:2604.24971 arXiv:2604.25724 arXiv:2604.25850 arXiv:2604.27032 arXiv:2605.00796 arXiv:2605.01280 arXiv:2605.01604 arXiv:2605.03310 arXiv:2605.05538 arXiv:2605.06647 arXiv:2605.06760 arXiv:2605.07234 arXiv:2605.08717 arXiv:2605.09635 arXiv:2605.10834 arXiv:2605.11202 arXiv:2605.11733 arXiv:2605.12493 arXiv:2605.15040 arXiv:2605.15957 arXiv:2605.16352 arXiv:2605.17613 arXiv:2605.18770 arXiv:2605.18825 arXiv:2605.19537 arXiv:2605.19743 arXiv:2605.19893 arXiv:2605.20173 arXiv:2605.20466 arXiv:2605.23950 arXiv:2605.24238 arXiv:2605.25475 arXiv:2605.25480 arXiv:2605.26112 arXiv:2605.27123 arXiv:2605.27492 arXiv:2605.27922 arXiv:2605.28120 arXiv:2605.28732 arXiv:2605.29640 arXiv:2605.30104 arXiv:2606.00610 arXiv:2606.01581 arXiv:2606.01613 arXiv:2606.01927 arXiv:2606.02522 arXiv:2606.02643 arXiv:2606.02964 arXiv:2606.05405 arXiv:2606.05608 arXiv:2606.06090 arXiv:2606.06252 arXiv:2606.07402 arXiv:2606.08340 arXiv:2606.09498 arXiv:2606.10953 arXiv:2606.13141 arXiv:2606.14589 arXiv:2606.16409 arXiv:2606.17107 arXiv:2606.17512 arXiv:2606.18037 arXiv:2606.19544 arXiv:2606.19787 arXiv:2606.20295 arXiv:2606.21649 arXiv:2606.24775 arXiv:2606.26458 arXiv:2606.26511 arXiv:2606.28070 arXiv:2606.31145 arXiv:2607.00406 arXiv:2607.00482 arXiv:2607.01774 arXiv:2607.02574 arXiv:2607.02588 arXiv:2607.04763 arXiv:2607.05196 arXiv:2607.05382 arXiv:2607.05391 arXiv:2607.05708 arXiv:2607.05876 arXiv:2607.06560 arXiv:2607.06815 arXiv:2607.06820 arXiv:2607.07050 arXiv:2607.07383 arXiv:2607.07702 arXiv:2607.07740 arXiv:2607.07858 arXiv:2607.07953 arXiv:2607.08028 arXiv:2607.08057 arXiv:2607.08269 arXiv:2607.08395 arXiv:2607.08404 arXiv:2607.08459 arXiv:2607.08495 arXiv:2607.08646 arXiv:2607.08662 arXiv:2607.08700 arXiv:2607.08716 arXiv:2607.08765 arXiv:2607.08964 arXiv:2607.08973 arXiv:2607.09092 arXiv:2607.09172 arXiv:2607.09322 arXiv:2607.09349 arXiv:2607.09424 arXiv:2607.09701 arXiv:2607.10183 arXiv:2607.10400 arXiv:2607.10428 arXiv:2607.10463 arXiv:2607.10522 arXiv:2607.11079 arXiv:2607.11111 arXiv:2607.11172 arXiv:2607.11250 arXiv:2607.11423 arXiv:2607.11433 arXiv:2607.11505 arXiv:2607.11523 arXiv:2607.11594 arXiv:2607.11643 arXiv:2607.11683 arXiv:2607.11706 arXiv:2607.11736 arXiv:2607.11783 arXiv:2607.11862 arXiv:2607.11881 arXiv:2607.11886 arXiv:2607.12227 arXiv:2607.12310 arXiv:2607.12340 arXiv:2607.12395 arXiv:2607.12406 arXiv:2607.12463 arXiv:2607.12477 arXiv:2607.12746 arXiv:2607.12747 arXiv:2607.12764 arXiv:2607.13027 arXiv:2607.13034 arXiv:2607.13104 arXiv:2607.13125 arXiv:2607.13276 arXiv:2607.13399 arXiv:2607.13705 arXiv:2607.13988 arXiv:2607.14076 arXiv:2607.14187 arXiv:2607.14277 arXiv:2607.14387 arXiv:2607.14541 arXiv:2607.14777 arXiv:2607.14811 arXiv:2607.14830 arXiv:2607.14952 arXiv:2607.15058 arXiv:2607.15095 arXiv:2607.15161 arXiv:2607.15207 arXiv:2607.15257 arXiv:2607.15263 arXiv:2607.15278 arXiv:2607.15434 arXiv:2607.15495 arXiv:2607.15657 arXiv:2607.15901 arXiv:2607.16165 arXiv:2607.16190 arXiv:2607.16617 arXiv:2607.16955 arXiv:2607.17247 arXiv:2607.17250 arXiv:2607.17524 arXiv:2607.17715 arXiv:2607.17986 arXiv:2607.18069 arXiv:2607.18141 arXiv:2607.18142 arXiv:2607.18144 arXiv:2607.18149 arXiv:2607.18155 arXiv:2607.18171 arXiv:2607.18195 arXiv:2607.18213 arXiv:2607.18529 arXiv:2607.18603 arXiv:2607.18754 arXiv:2607.18772 arXiv:2607.18825 arXiv:2607.19011 arXiv:2607.19215 arXiv:2607.19238 arXiv:2607.19297 arXiv:2607.19604 arXiv:2607.19636 arXiv:2607.19747 arXiv:2607.19793 arXiv:2607.19865 arXiv:2607.19867 arXiv:2607.20061 arXiv:2607.20145 arXiv:2607.20346 arXiv:2607.20368 arXiv:2607.20465 arXiv:2607.20709 arXiv:2607.20734 arXiv:2607.20785 arXiv:2607.20868 arXiv:2607.20891 arXiv:2607.20911 arXiv:2607.20957 arXiv:2607.21051 arXiv:2607.21072 arXiv:2607.21324 arXiv:2607.21461 arXiv:2607.21485 arXiv:2607.21503 arXiv:2607.21553 arXiv:2607.21556 arXiv:2607.21557 arXiv:2607.21576 arXiv:2607.21596 arXiv:2607.21653 arXiv:2607.21655 arXiv:2607.21694 arXiv:2607.21962 arXiv:2607.22043 arXiv:2607.22091 arXiv:2607.22098 arXiv:2607.22157 arXiv:2607.22375 arXiv:2607.22389 arXiv:2607.22529 arXiv:2607.22561 arXiv:2607.22688 arXiv:2607.22724 arXiv:2607.22798 arXiv:2607.23193 arXiv:2607.23242 arXiv:2607.23263 arXiv:2607.23402 arXiv:2607.23514 arXiv:2607.23518 arXiv:2607.23588 arXiv:2607.23802 arXiv:2607.23855 arXiv:2607.23909 arXiv:2607.24000 arXiv:2607.24027 arXiv:2607.24062 arXiv:2607.24117 arXiv:2607.24223 arXiv:2607.24280 arXiv:2607.24352 arXiv:2607.24368 arXiv:2607.24475 arXiv:2607.24554 arXiv:2607.24651 arXiv:2607.24653 arXiv:2607.24663 arXiv:2607.24720 arXiv:2607.24731 arXiv:2607.24744 arXiv:2607.24882 arXiv:2607.24904 arXiv:2607.24957 arXiv:2607.25236 arXiv:2607.25294 arXiv:2607.25308 arXiv:2607.25337 arXiv:2607.25379 arXiv:2607.25380 arXiv:2607.25398 arXiv:2607.25431 arXiv:2607.25498 arXiv:2607.25537 arXiv:2607.25565 arXiv:2607.25572 arXiv:2607.25600 arXiv:2607.25659 arXiv:2607.25675 arXiv:2607.25857 arXiv:2607.25895 arXiv:2607.25959 arXiv:2607.25996 arXiv:2607.26004 arXiv:2607.26037 arXiv:2607.26055 arXiv:2607.26057 arXiv:2607.26314 arXiv:2607.26410 arXiv:2607.26451 arXiv:2607.26475 arXiv:2607.26497 arXiv:2607.26520 arXiv:2607.26571 arXiv:2607.26637 arXiv:2607.26654 arXiv:2607.26760 arXiv:2607.26769 arXiv:2607.26784 arXiv:2607.27090 arXiv:2607.27146 arXiv:2607.27167 arXiv:2607.27180 arXiv:2607.27201 arXiv:2607.27205 arXiv:2607.27506 arXiv:2607.27600 arXiv:2607.27616 arXiv:2607.27816 arXiv:2607.27851 arXiv:2607.27918 arXiv:2607.27919 arXiv:2607.27958 arXiv:2607.28074 arXiv:2607.28126 arXiv:2607.28227 arXiv:2607.28229 arXiv:2607.28263 arXiv:2607.28319 arXiv:2607.28362 arXiv:2607.28374 arXiv:2607.28397 arXiv:2607.28509 arXiv:2607.28580 arXiv:2607.28595 arXiv:2607.28609 arXiv:2607.28617 arXiv:2607.28618 arXiv:2607.28633 arXiv:2607.28675 arXiv:2607.28802 arXiv:2607.29007 arXiv:2607.29167 arXiv:2607.29402 arXiv:2607.29459 arXiv:2607.29591 arXiv:2607.29610 arXiv:2607.29677 arXiv:2607.29678 arXiv:2608.00267 arXiv:2608.00650 arXiv:2608.00675 arXiv:2608.00677 arXiv:2608.00730 arXiv:2608.00881 arXiv:2608.00902 arXiv:2608.00922 arXiv:2608.01247 arXiv:2608.01526 arXiv:2608.01651 arXiv:2608.01678 arXiv:2608.01735 arXiv:2608.01964 arXiv:2608.02218 arXiv:2608.02583 arXiv:2608.02703 arXiv:2608.02713 arXiv:2608.02870 arXiv:2608.02989 arXiv:2608.03451 arXiv:2608.03499 arXiv:2608.03506 arXiv:2608.03756 arXiv:2608.03764 arXiv:2608.03874 arXiv:2608.03887 arXiv:2608.03893 arXiv:2608.03979 arXiv:2608.04003 arXiv:2608.04530 arXiv:2608.04569 arXiv:2608.04964 arXiv:2608.05102 arXiv:2608.05138 arXiv:2608.05604 arXiv:2608.05703 arXiv:2608.05747 arXiv:2608.05784 arXiv:2608.05959 arXiv:2608.06020 arXiv:2608.06033 arXiv:2608.06130 arXiv:2608.06146 arXiv:2608.06197 arXiv:2608.06216 arXiv:2608.06301 arXiv:2608.06312 arXiv:2608.06329 arXiv:2608.06352 arXiv:2608.06729 arXiv:2608.06790 arXiv:2608.06867 arXiv:2608.07009 arXiv:2608.07152 arXiv:2608.07370 arXiv:2608.07446 arXiv:2608.07458 arXiv:2608.07545 arXiv:2608.07645 arXiv:2608.08020 arXiv:2608.08097 arXiv:2608.08119 arXiv:2608.08160 arXiv:2608.08311 arXiv:2608.08466 arXiv:2608.08621 arXiv:2608.08676 arXiv:2608.08722 arXiv:2608.08814 arXiv:2608.08975 arXiv:2608.09096 arXiv:2608.09158 arXiv:2608.09253 arXiv:2608.09408 arXiv:2608.09442 arXiv:2608.09766 arXiv:2608.09779 arXiv:2608.09802 arXiv:2608.09867 arXiv:2608.09888 arXiv:2608.09900 arXiv:2608.10288 arXiv:2608.10299 arXiv:2608.10450 arXiv:2608.10708 arXiv:2608.10720 arXiv:2608.10744 arXiv:2608.10875 arXiv:2608.11030 arXiv:2608.11205 arXiv:2608.11367 arXiv:2608.11632 arXiv:2608.11660 arXiv:2608.11745 arXiv:2608.11924 arXiv:2608.12036 arXiv:2608.12149 arXiv:2608.12184 arXiv:2608.12218 arXiv:2608.12253 arXiv:2608.12304 arXiv:2608.12307 arXiv:2608.12313 arXiv:2608.12314 arXiv:2608.12440 arXiv:2608.12564 arXiv:2608.12627 arXiv:2608.12743 arXiv:2608.12781 arXiv:2608.12812 arXiv:2608.12875 arXiv:2608.12888 arXiv:2608.13010 arXiv:2608.13120 arXiv:2608.13237 arXiv:2608.13410 arXiv:2608.13489 arXiv:2608.13505 arXiv:2608.13545 arXiv:2608.13546 arXiv:2608.13547 arXiv:2608.13552 arXiv:2608.13558 arXiv:2608.13560 arXiv:2608.13622 arXiv:2608.13867 arXiv:2608.13947 arXiv:2608.14036 arXiv:2608.14229 arXiv:2608.14333 arXiv:2608.14376 arXiv:2608.14380 arXiv:2608.14929 arXiv:2608.15089 arXiv:2608.15380 arXiv:2608.15669 arXiv:2608.15767 arXiv:2608.15875 arXiv:2608.15888 arXiv:2608.16072 arXiv:2608.16157 arXiv:2608.16590 arXiv:2608.16628 arXiv:2608.16859 arXiv:2608.16885 arXiv:2608.16977 arXiv:2608.17050 arXiv:2608.17067 arXiv:2608.17253 arXiv:2608.17271 arXiv:2608.17310 arXiv:2608.17379 arXiv:2608.17393 arXiv:2608.17402 arXiv:2608.17423 arXiv:2608.17426 arXiv:2608.17528 arXiv:2608.17536 arXiv:2608.17597 arXiv:2608.17781 arXiv:2608.17950 arXiv:2608.17960 arXiv:2608.18027 arXiv:2608.18063 arXiv:2608.18077 arXiv:2608.18092 arXiv:2608.18184 arXiv:2608.18484 arXiv:2608.18489 arXiv:2608.18524 arXiv:2608.18565 arXiv:2608.18607 arXiv:2608.18613 arXiv:2608.18682 arXiv:2608.18746 arXiv:2608.18852 arXiv:2608.19197 arXiv:2608.19269 arXiv:2608.19652 arXiv:2608.19662 arXiv:2608.19758 arXiv:2608.19799 arXiv:2608.19854 arXiv:2608.19857 arXiv:2608.19880 arXiv:2608.19936 arXiv:2608.20055 arXiv:2608.20169 arXiv:2608.20202 arXiv:2608.20246 arXiv:2608.20281 arXiv:2608.20317 arXiv:2608.20318 arXiv:2608.20319 arXiv:2608.20331 arXiv:2608.20335 arXiv:2608.20338 arXiv:2608.20438 arXiv:2608.20574 arXiv:2608.20953 arXiv:2608.21031 arXiv:2608.21095 arXiv:2608.21140 arXiv:2608.21159 arXiv:2608.21208 arXiv:2608.21249 arXiv:2608.21252 arXiv:2608.21281 arXiv:2608.21360 arXiv:2608.21450 arXiv:2608.21500 arXiv:2608.21832 arXiv:2608.21833 arXiv:2608.21839 arXiv:2608.22339 arXiv:2608.22505 arXiv:2608.22510 arXiv:2608.22752 arXiv:2608.22872 arXiv:2608.23035 arXiv:2608.23200 arXiv:2608.23252 arXiv:2608.23552 arXiv:2608.23564 arXiv:2608.23670 arXiv:2608.23691 arXiv:2608.23740 arXiv:2608.24040 arXiv:2608.24053 arXiv:2608.24358 arXiv:2608.24777 arXiv:2608.24804 arXiv:2608.25417 arXiv:2608.25500 arXiv:2608.25518 arXiv:2608.25593 arXiv:2608.25625 arXiv:2608.25735 arXiv:2608.25986 arXiv:2608.26005 arXiv:2608.26070 arXiv:2608.26091 arXiv:2608.26131 arXiv:2608.26200 arXiv:2608.26238 arXiv:2608.26530 arXiv:2608.26582 arXiv:2608.26623 arXiv:2608.26730 arXiv:2608.26794 arXiv:2608.26809 arXiv:2608.26836 arXiv:2608.26872 arXiv:2608.26993 arXiv:2608.27260 arXiv:2608.27345 arXiv:2608.27448 arXiv:2608.27455 arXiv:2608.27456 arXiv:2608.27529 arXiv:2608.27809 arXiv:2608.27875 arXiv:2608.27969 arXiv:2608.28122 arXiv:2608.28165 arXiv:2608.28281 arXiv:2608.28389 arXiv:2608.28444 arXiv:2608.28460 arXiv:2608.28695 arXiv:2608.29188 arXiv:2608.29310 arXiv:2608.29387 arXiv:2608.29464 arXiv:2608.29607 arXiv:2608.30005 arXiv:2608.30322 arXiv:2608.30391 arXiv:2608.30478 arXiv:2608.30730 arXiv:2608.31022 arXiv:2608.31082 arXiv:2608.31100 arXiv:2608.31111 arXiv:2608.31139 arXiv:2609.00006 arXiv:2609.00092 arXiv:2609.00137 arXiv:2609.00196 arXiv:2609.00224 arXiv:2609.00365 arXiv:2609.00474 arXiv:2609.00551 arXiv:2609.01072 arXiv:2609.01245 arXiv:2609.01437 arXiv:2609.01481 arXiv:2609.01507 arXiv:2609.01572 arXiv:2609.01601 arXiv:2609.01657 arXiv:2609.01836 arXiv:2609.01925 arXiv:2609.02367 arXiv:2609.02486 arXiv:2609.02745 arXiv:2609.02750 arXiv:2609.02783 arXiv:2609.02859 arXiv:2609.02998 arXiv:2609.03153 arXiv:2609.03199 arXiv:2609.03241 arXiv:2609.03293 arXiv:2609.03430 arXiv:2609.03586 arXiv:2609.03756 arXiv:2609.03796 arXiv:2609.03820 arXiv:2609.04010 arXiv:2609.04034 arXiv:2609.04043 arXiv:2609.04061 arXiv:2609.04094 arXiv:2609.04098 arXiv:2609.04131 arXiv:2609.04148 arXiv:2609.04170 arXiv:2609.04172 arXiv:2609.04173 arXiv:2609.04196 arXiv:2609.04199 arXiv:2609.04201 arXiv:2609.04250 arXiv:2609.04280 arXiv:2609.04304 arXiv:2609.04382 arXiv:2609.04438 arXiv:2609.04444 arXiv:2609.04482 arXiv:2609.04490 arXiv:2609.04523 arXiv:2609.04611 arXiv:2609.04714 arXiv:2609.04720 arXiv:2609.04753 arXiv:2609.05095 arXiv:2609.05110 arXiv:2609.05232 arXiv:2609.05258 arXiv:2609.05295 arXiv:2609.05339 arXiv:2609.05405 arXiv:2609.05415 arXiv:2609.05416 arXiv:2609.05571 arXiv:2609.05661 arXiv:2609.05736 arXiv:2609.05760 arXiv:2609.05903 arXiv:2609.06011 arXiv:2609.06107 arXiv:2609.06140 arXiv:2609.06245 arXiv:2609.06674 arXiv:2609.06702 arXiv:2609.06986 arXiv:2609.07064 arXiv:2609.07395 arXiv:2609.07529 arXiv:2609.08126 arXiv:2609.08149 arXiv:2609.08183 arXiv:2609.08365 arXiv:2609.08368 arXiv:2609.08418 arXiv:2609.08572 arXiv:2609.08832 arXiv:2609.08936 arXiv:2609.08950 arXiv:2609.08977 arXiv:2609.09002 arXiv:2609.09113 arXiv:2609.09134 arXiv:2609.09155 arXiv:2609.09219 arXiv:2609.09875 arXiv:2609.10016 arXiv:2609.10226 arXiv:2609.10266 arXiv:2609.10522 arXiv:2609.10712 arXiv:2609.10715 arXiv:2609.10745 arXiv:2609.10895 arXiv:2609.11042 arXiv:2609.11085 arXiv:2609.11108 arXiv:2609.11115 arXiv:2609.11294 arXiv:2609.11317 arXiv:2609.11390 arXiv:2609.11412 arXiv:2609.11486 arXiv:2609.11561 arXiv:2609.11596 arXiv:2609.11638 arXiv:2609.11646 arXiv:2609.11699 arXiv:2609.11808 arXiv:2609.11873 arXiv:2609.11929 arXiv:2609.11977 arXiv:2609.12397 arXiv:2609.12641 arXiv:2609.13141 arXiv:2609.13285 arXiv:2609.13356 arXiv:2609.13406 arXiv:2609.13463 arXiv:2609.13814 arXiv:2609.14005 arXiv:2609.14022 arXiv:2609.14302 arXiv:2609.14320 arXiv:2609.14462 arXiv:2609.14858 arXiv:2609.14973 arXiv:2609.15029 arXiv:2609.15126 arXiv:2609.15134 arXiv:2609.15195 arXiv:2609.15364 arXiv:2609.15478 arXiv:2609.15504 arXiv:2609.15578 arXiv:2609.15659 arXiv:2609.15779 arXiv:2609.15800 arXiv:2609.15810 arXiv:2609.15818 arXiv:2609.15830 arXiv:2609.15863 arXiv:2609.15938 arXiv:2609.15972 arXiv:2609.15973 arXiv:2609.15987 arXiv:2609.16034 arXiv:2609.16409 arXiv:2609.16453 arXiv:2609.16591 arXiv:2609.16679 arXiv:2609.16816 arXiv:2609.16818 arXiv:2609.16900 arXiv:2609.17012 arXiv:2609.17488 arXiv:2609.17496 arXiv:2609.17523 arXiv:2609.17524 arXiv:2609.17632 arXiv:2609.17652 arXiv:2609.17653 arXiv:2609.17708 arXiv:2609.17909 arXiv:2609.18063 arXiv:2609.18094 arXiv:2609.18323 arXiv:2609.18487 arXiv:2609.18605 arXiv:2609.18703 arXiv:2609.18708 arXiv:2609.18779 arXiv:2609.18805 arXiv:2609.19134 arXiv:2609.19138 arXiv:2609.19499 arXiv:2609.19656 arXiv:2609.19671 arXiv:2609.19879 arXiv:2609.19969 arXiv:2609.20423 arXiv:2609.20511 arXiv:2609.20519 arXiv:2609.20715 arXiv:2609.20744 arXiv:2609.20784 arXiv:2609.20800 arXiv:2609.20804 arXiv:2609.20817 arXiv:2609.21346 arXiv:2609.21465 arXiv:2609.22000 arXiv:2609.22068 arXiv:2609.22076 arXiv:2609.22319 arXiv:2609.22682 arXiv:2609.22753 arXiv:2609.22947 arXiv:2609.23038 arXiv:2609.23087 arXiv:2609.23088 arXiv:2609.23407 arXiv:2609.23551 arXiv:2609.23863 arXiv:2609.23989 arXiv:2609.24058 arXiv:2609.24220 arXiv:2609.24308 arXiv:2609.24385 arXiv:2609.24555 arXiv:2609.24657 arXiv:2609.24749 arXiv:2609.24815 arXiv:2609.24967 arXiv:2609.24981 arXiv:2609.24983 arXiv:2609.24997 arXiv:2609.25001 arXiv:2609.25053 arXiv:2609.25165 arXiv:2609.25186 arXiv:2609.25187 arXiv:2609.25270 arXiv:2609.25716 arXiv:2609.25804 arXiv:2609.25853 arXiv:2609.25963 arXiv:2609.25991 arXiv:2609.26072 arXiv:2609.26333 arXiv:2609.26355 arXiv:2609.26489 arXiv:2609.26550 arXiv:2609.26634 arXiv:2609.26637 arXiv:2609.26774 arXiv:2609.26777 arXiv:2609.26780 arXiv:2609.26781 arXiv:2609.27158 arXiv:2609.27213 arXiv:2609.27277 arXiv:2609.27308 arXiv:2609.27334 arXiv:2609.27657 arXiv:2609.27901 arXiv:2609.27980 arXiv:2609.28327 arXiv:2609.28416 arXiv:2609.28466 arXiv:2609.28470 arXiv:2609.28473 arXiv:2609.28603 arXiv:2609.28654 arXiv:2609.28811 arXiv:2609.28845 arXiv:2609.28923 arXiv:2609.29028 arXiv:2609.29362 arXiv:2609.29421 arXiv:2609.29429 arXiv:2609.29444 arXiv:2609.29647 arXiv:2609.29695 arXiv:2609.29803 arXiv:2609.29815 arXiv:2609.29816 arXiv:2609.29837 arXiv:2609.29845 arXiv:2609.29964 arXiv:2609.29973 arXiv:2609.29983 arXiv:2609.30167 arXiv:2609.30184 arXiv:2609.30192 arXiv:2609.30216 arXiv:2609.30221 arXiv:2609.30222 arXiv:2609.30233 arXiv:2609.30238 arXiv:2609.30243 arXiv:2609.30250 arXiv:2609.30904 arXiv:2609.31002 arXiv:2609.31093 arXiv:2609.31291 arXiv:2609.31394 arXiv:2609.31397 arXiv:2609.31415 arXiv:2609.31590 arXiv:2609.31620 arXiv:2609.31847 arXiv:2609.32049 arXiv:2609.32019 arXiv:2609.32259 arXiv:2609.32264 arXiv:2609.32522 arXiv:2609.32577 arXiv:2609.32600 arXiv:2609.32607 arXiv:2609.32720 arXiv:2609.32722 arXiv:2609.32759 arXiv:2609.32810 arXiv:2609.32921 arXiv:2609.32965 arXiv:2609.32993 arXiv:2609.33355 arXiv:2609.33382 arXiv:2609.33485 arXiv:2609.33591 arXiv:2609.33697 arXiv:2609.33780 arXiv:2609.33987 arXiv:2609.34242 arXiv:2609.34262 arXiv:2609.34274 arXiv:2609.34385 arXiv:2609.34392 arXiv:2609.34563 arXiv:2609.34645 arXiv:2609.34826 arXiv:2609.34895 arXiv:2609.35215 arXiv:2609.35259 arXiv:2609.35427 arXiv:2609.35530 arXiv:2609.35629 arXiv:2609.35673 arXiv:2609.35718 arXiv:2609.35767 arXiv:2609.35932 arXiv:2609.36071 arXiv:2609.36138 arXiv:2609.36246 arXiv:2609.36314 arXiv:2609.36322 arXiv:2609.36388 arXiv:2609.36435 arXiv:2609.36550 arXiv:2609.36585 arXiv:2609.36636 arXiv:2609.36651 arXiv:2609.36730 arXiv:2609.36820 arXiv:2609.36965 arXiv:2609.37147 arXiv:2609.37226 arXiv:2609.37236 arXiv:2609.37533 arXiv:2609.37559 arXiv:2609.37673 arXiv:2609.37686 arXiv:2609.37687 arXiv:2609.37690 arXiv:2609.37725 arXiv:2609.37749 arXiv:2609.37863 arXiv:2609.37925 arXiv:2609.37969 arXiv:2609.38096 arXiv:2609.38123 arXiv:2609.38137 arXiv:2609.38334 arXiv:2609.38349 arXiv:2609.38426 arXiv:2609.38527 arXiv:2609.38653 arXiv:2609.38660 arXiv:2609.38721 arXiv:2609.38792 arXiv:2609.38879 arXiv:2609.38987 arXiv:2609.39068 arXiv:2609.39071 arXiv:2609.39075 arXiv:2609.39096 arXiv:2609.39102 arXiv:2609.39154 arXiv:2609.39358 arXiv:2609.39378 arXiv:2609.39420 arXiv:2609.39490 arXiv:2609.39788 arXiv:2609.39841 arXiv:2609.39903 arXiv:2609.39929 arXiv:2609.39982 arXiv:2609.40222 arXiv:2609.40247 arXiv:2609.40316 arXiv:2609.40362 arXiv:2610.00437 arXiv:2610.00544 arXiv:2610.00559 arXiv:2610.00574 arXiv:2610.00722 arXiv:2610.00812 arXiv:2610.00972 arXiv:2610.01092 arXiv:2610.01139 arXiv:2610.01257 arXiv:2610.01324 arXiv:2610.01393 arXiv:2610.01490 arXiv:2610.01553 arXiv:2610.01664 arXiv:2610.01705 arXiv:2610.01762 arXiv:2610.01767 arXiv:2610.01780 arXiv:2610.01871 arXiv:2610.01936 arXiv:2610.01939 arXiv:2610.02076 arXiv:2610.02122 arXiv:2610.02150 arXiv:2610.02185 arXiv:2610.02188 arXiv:2610.02191 arXiv:2610.02196 arXiv:2610.02201 arXiv:2610.02206 arXiv:2610.02304 arXiv:2610.02381 arXiv:2610.02396 arXiv:2610.02508 arXiv:2610.02521 arXiv:2610.02710 arXiv:2610.02772 arXiv:2610.02832 arXiv:2610.02840 arXiv:2610.03020 arXiv:2610.03120 arXiv:2610.03136 arXiv:2610.03140 arXiv:2610.03240 arXiv:2610.03367 arXiv:2610.03391 arXiv:2610.03421 arXiv:2610.03509 arXiv:2610.03574 arXiv:2610.03632 arXiv:2610.03664 arXiv:2610.03665 arXiv:2610.03710 arXiv:2610.03715 arXiv:2610.04116 arXiv:2610.04596 arXiv:2610.04616 arXiv:2610.04631 arXiv:2610.04722 arXiv:2610.04749 arXiv:2610.05033 arXiv:2610.05162 arXiv:2610.05622 arXiv:2610.05709 arXiv:2610.05782 arXiv:2610.05826 arXiv:2610.05842 arXiv:2610.05912 arXiv:2610.05949 arXiv:2610.06184 arXiv:2610.06207 arXiv:2610.06479 arXiv:2610.06666 arXiv:2610.06679 arXiv:2610.06830 arXiv:2610.06844 arXiv:2610.07250 arXiv:2610.07332 arXiv:2610.07384 arXiv:2610.07557 arXiv:2610.07753 arXiv:2610.08048 arXiv:2610.08077 arXiv:2610.08244 arXiv:2610.08432 arXiv:2610.08452 arXiv:2610.08463 arXiv:2610.08571 arXiv:2610.08630 arXiv:2610.08674 arXiv:2610.08902 arXiv:2610.08995 arXiv:2610.09228 arXiv:2610.09280 arXiv:2610.09684 arXiv:2610.09832 arXiv:2610.10091 arXiv:2610.10507 arXiv:2610.10533 arXiv:2610.10845 arXiv:2610.11169 arXiv:2610.11287 arXiv:2610.11794 arXiv:2610.11899 arXiv:2610.11959 arXiv:2610.12085 arXiv:2610.12274 arXiv:2610.12299 arXiv:2610.12312 arXiv:2610.12341 arXiv:2610.12415 arXiv:2610.12419 arXiv:2610.12421 arXiv:2610.12423 arXiv:2610.12458 arXiv:2610.12461 arXiv:2610.12611 arXiv:2610.13735 arXiv:2609.39600 arXiv:2610.09116 arXiv:2610.13166 arXiv:2610.13361 arXiv:2602.06176 arXiv:2602.09305 arXiv:2604.15149 arXiv:2609.34227 arXiv:2609.37334 arXiv:2609.38078
7.2 完整 URL 列表(v121 沿用 395 件 + v122 net-new 8 件 + v123 net-new 7 件 · 418 件总览)
https://academy.dair.ai/papers/looparena-benchmarking-models-as-runtime-controllers-for-loop-engineering-2608.28281 https://aembit.io/blog/the-ultimate-guide-to-mcp-security-vulnerabilities https://aiamastery.substack.com https://aimultiple.com/agentic-llm https://aishwaryasrinivasan.substack.com/p/all-you-need-to-know-about-harness https://alice-labs.com/best-ai-agent-frameworks-2026 https://alphasignal.substack.com/p/rag-and-long-context-arent-enough https://alphasignalai.substack.com/p/rag-and-long-context-arent-enough https://alphaxiv.org/abs/2609.00196 https://arxiv.org/abs/2609.02783 https://arxiv.org/abs/2609.16591 https://amap-ml.github.io/LoopArena https://anthropic.com/news/discovering-cryptographic-weaknesses-with-claude https://anthropic.com/news/three-real-world-incidents-cybersecurity-evaluations https://app.opencve.io/cve https://arxiv.org/abs/2509.23040 https://arxiv.org/abs/2510.05381 https://arxiv.org/abs/2510.09665 https://arxiv.org/abs/2510.15253 https://arxiv.org/abs/2511.01815 https://arxiv.org/abs/2511.16681 https://arxiv.org/abs/2601.03236 https://arxiv.org/abs/2601.06112 https://arxiv.org/abs/2602.21548 https://arxiv.org/abs/2602.23368 https://arxiv.org/abs/2603.00873 https://arxiv.org/abs/2604.06185 https://arxiv.org/abs/2604.25724 https://arxiv.org/abs/2605.01495 https://arxiv.org/abs/2605.03310 https://arxiv.org/abs/2605.15040 https://arxiv.org/abs/2605.18825 https://arxiv.org/abs/2606.01613 https://arxiv.org/abs/2606.06090 https://arxiv.org/abs/2606.17107 https://arxiv.org/abs/2606.18037 https://arxiv.org/abs/2606.21649 https://arxiv.org/abs/2606.26511 https://arxiv.org/abs/2607.08057 https://arxiv.org/abs/2607.08716 https://arxiv.org/abs/2608.06790 https://arxiv.org/abs/2608.12564 https://arxiv.org/abs/2608.13867 https://arxiv.org/abs/2608.25735 https://arxiv.org/abs/2608.26131 https://arxiv.org/abs/2608.26836 https://arxiv.org/abs/2608.27809 https://arxiv.org/abs/2608.28389 https://arxiv.org/abs/2609.08365 https://arxiv.org/abs/2609.11873 https://arxiv.org/abs/2609.13406 https://arxiv.org/abs/2609.15504 https://arxiv.org/abs/2609.15578 https://arxiv.org/abs/2609.15972 https://arxiv.org/abs/2609.16453 https://arxiv.org/abs/2609.16818 https://arxiv.org/abs/2609.17012 https://arxiv.org/abs/2609.17524 https://arxiv.org/abs/2609.24749 https://arxiv.org/abs/2609.26333 https://arxiv.org/abs/2609.30192 https://arxiv.org/abs/2609.31415 https://arxiv.org/abs/2609.31847 https://arxiv.org/abs/2609.32049 https://arxiv.org/abs/2609.32577 https://arxiv.org/abs/2609.32965 https://arxiv.org/abs/2609.33591 https://arxiv.org/abs/2609.34242 https://arxiv.org/abs/2609.34385 https://arxiv.org/abs/2609.35215 https://arxiv.org/abs/2609.35629 https://arxiv.org/abs/2609.35932 https://arxiv.org/abs/2609.36965 https://arxiv.org/abs/2609.37226 https://arxiv.org/abs/2609.37725 https://arxiv.org/abs/2609.37749 https://arxiv.org/abs/2609.40222 https://arxiv.org/abs/2609.40362 https://arxiv.org/abs/2609.41412 https://arxiv.org/abs/2610.00437 https://arxiv.org/abs/2610.00544 https://arxiv.org/abs/2610.00559 https://arxiv.org/abs/2610.00574 https://arxiv.org/abs/2610.00812 https://arxiv.org/abs/2610.01092 https://arxiv.org/abs/2610.01762 https://arxiv.org/abs/2610.01767 https://arxiv.org/abs/2610.01871 https://arxiv.org/abs/2610.01936 https://arxiv.org/abs/2610.01939 https://arxiv.org/abs/2610.02122 https://arxiv.org/abs/2610.02185 https://arxiv.org/abs/2610.02196 https://arxiv.org/abs/2610.02201 https://arxiv.org/abs/2609.33987 https://arxiv.org/abs/2610.11169 https://arxiv.org/abs/2610.11794 https://arxiv.org/abs/2610.12299 https://arxiv.org/abs/2610.12419 https://arxiv.org/abs/2608.20055 https://arxiv.org/abs/2610.11899 https://arxiv.org/abs/2610.12085 https://arxiv.org/abs/2610.12312 https://arxiv.org/abs/2610.12415 https://arxiv.org/abs/2610.02396 https://arxiv.org/abs/2610.08902 https://arxiv.org/abs/2610.08077 https://arxiv.org/abs/2610.12458 https://arxiv.org/html/2501.09136v4 https://arxiv.org/html/2509.23040v5 https://arxiv.org/html/2510.05381 https://arxiv.org/html/2601.07711v1 https://arxiv.org/html/2602.03442v1 https://arxiv.org/html/2605.15040 https://arxiv.org/html/2606.14589 https://arxiv.org/html/2607.08716v1 https://arxiv.org/html/2608.06301 https://arxiv.org/html/2608.06790 https://arxiv.org/pdf/2310.11703 https://arxiv.org/pdf/2608.26131 https://arxiv.org/pdf/2610.12299v1 https://ascendai.csdn.net/69d4c85172111d255bf7caad.html https://blog.bytebytego.com/p/ep227-top-9-places-to-use-jev https://blog.bytebytego.com/p/how-chatgpt-optimizes-its-agent-loop https://blog.bytebytego.com/p/how-openai-built-gpt-live https://blog.cloudflare.com/p/agent-access-model https://blog.csdn.net/2501_93100691/article/details/162990807 https://blog.csdn.net/weixin_41870061/article/details/162933242 https://blog.csdn.net/xx_nm98/article/details/158851692 https://blog.csdn.net/xxn_m98/article/details/158851692 https://blog.gitguardian.com/hugging-face-breach-ai-agent-security-lessons https://buzzgrewal.medium.com/ai-agents-dont-need-vector-search-anymore-inside-the-agentic-search-stack-replacing-rag-in-2026-58efcabe4f6f https://cameronrwolfe.substack.com/p/agentic-rl https://cameronrwolfe.substack.com/p/agentic-world-models https://cameronrwolfe.substack.com/p/comprehensive-guide-agent-evaluation https://cameronrwolfe.substack.com/p/midtraining https://chatpaper.com/chatpaper/paper/241604 https://chatpaper.com/paper/241604 https://chatpaper.com/paper/337816 https://chatpaper.com/paper/341600 https://christian-schneider.net/blog/rag-security-forgotten-attack-surface https://claudiostamile.substack.com/p/agent-memory-is-not-rag-a-practical https://cockroachlabs.com/blog/a2a-agent-state-data-layer/ https://codingwithroby.substack.com/p/the-2026-ai-agent-stack-drawn-from https://darioamodei.com/post/we-must-pace-the-frontier https://dblp.org/rec/journals/corr/abs-2602-23368 https://dev.to/reidmarlow/why-coding-agents-fail-in-the-outer-loop-4c0f https://developers.googleblog.com/scaling-ai-agent-infrastructure-with-the-mcp-stateless-updates https://docs.aws.amazon.com/sagemaker/latest/dg/monitoring-cloudwatch-detailed-observability.html https://emergentmind.com/topics/harnessopt https://explainx.ai/blog/perplexity-q2d-web-agentic-rag-benchmark-2026 https://forbes.com/sites/janakirammsv/2026/08/19/trueforge https://fortune.com/2026/09/12/sam-altman-openai-ipo-2026/ https://futureagi.com/blog/llm-incident-response-playbook-2026 https://futureagi.substack.com https://futureagi.substack.com/p/the-llm-incident-runbook-six-steps-f27 https://garymarcus.substack.com/p/two-cheers-out-of-three-for-dario https://github.com/AMAP-ML/LoopArena https://github.com/AMAP-ML/LongHorizon-Harness https://github.com/Andrew0613/PAWBench https://github.com/JustVugg/colibri https://github.com/LLMSecurity/awesome-agent-skills-security https://github.com/LukaszGrochal/agent-framework-benchmark https://github.com/SensenGao/Multimodal-RAG-Survey-For-Document https://github.com/TencentCloud/TencentDB-Agent-Memory https://github.com/UKGovernmentBEIS/inspect_evals https://github.com/YennNing/MC-Search https://github.com/Yigtwxx/Awesome-RAG-Production https://github.com/ai-boost/awesome-harness-engineering https://github.com/cyberlife-coder/VelesDB https://github.com/epoch-research/MirrorCode https://github.com/google-research/envharness https://github.com/karpathy/autoresearch https://github.com/karpathy/nanochat https://github.com/lyogavin/airllm https://github.com/microsoft-foundry/discussions/177 https://github.com/microsoft/Echoverse https://github.com/microsoft/Orchard https://github.com/microsoft/SymCrypt/tree/feature/verifiedcrypto https://github.com/microsoft/ai-agents-for-beginners https://github.com/microsoft/autogen https://github.com/microsoft/orchard https://github.com/neulab/think-before-you-link https://github.com/sgl-project/sglang https://github.com/simonw/datasette-mcp https://github.com/simonw/mcp-explorer https://github.com/sjarmak/engineering-reliable-coding-agents https://github.com/syr-cn/ReMemR1 https://github.com/truefoundry/trueforge https://github.com/ulab-uiuc/AgentDebug https://github.com/umwyf/CRITICL https://github.com/volcengine/OpenViking https://github.com/xiyuanyang45/WMRL https://gradientflow.com/the-ai-data-center-backlash-has-a-blind-spot/ https://gradientflow.com/the-biggest-ai-risks-sit-outside-the-model/ https://hodgesj.substack.com/p/zero-to-hero-benchmarksdatasets-for https://huggingface.co/Accio-Lab/occamy-1.0 https://huggingface.co/TokenRhythm/NeoHorse-1-9B https://huggingface.co/blog/agent-intrusion-technical-timeline https://huggingface.co/blog/security-incident-july-2026 https://huggingface.co/datasets/AIOR-Research/OR-Clarify https://huggingface.co/datasets/Andrew613/PAWBench https://huggingface.co/datasets/microsoft/Echoverse https://huggingface.co/datasets/microsoft/Orchard https://huggingface.co/datasets/microsoft/orchard https://huggingface.co/datasets/neulab/merlin-rare https://huggingface.co/papers/2601.06112 https://huggingface.co/papers/2605.15040 https://huggingface.co/papers/2607.08716 https://huggingface.co/papers/2608.12564 https://huggingface.co/papers/2608.18524 https://huggingface.co/papers/2608.25518 https://huggingface.co/papers/2608.26131 https://huggingface.co/papers/2608.28122 https://huggingface.co/papers/2608.28281 https://huggingface.co/papers/2608.29310 https://huggingface.co/papers/2608.29464 https://huggingface.co/papers/2609.00196 https://huggingface.co/papers/2609.02783 https://huggingface.co/papers/2609.04201 https://huggingface.co/papers/2609.11977 https://huggingface.co/papers/2609.16591 https://huggingface.co/papers/2609.24749 https://huggingface.co/papers/2609.26333 https://huggingface.co/papers/2609.31847 https://huggingface.co/papers/2609.32577 https://huggingface.co/papers/2609.32965 https://huggingface.co/papers/2609.34242 https://huggingface.co/papers/2609.34385 https://huggingface.co/papers/2609.36965 https://huggingface.co/papers/2609.37226 https://huggingface.co/papers/2609.37725 https://huggingface.co/papers/2609.37749 https://huggingface.co/papers/2610.12299 https://iclr.cc/virtual/2026/poster/10011811 https://icml.cc/virtual/2026/events/2026-position-papers https://icml.cc/virtual/2026/poster/66210 https://importai.substack.com/p/import-ai-469-science-ai-rsi-simulator https://importai.substack.com/p/import-ai-472-deepminds-cheating https://importai.substack.com/p/import-ai-473-the-uss-superintelligence https://kili-technology.com/blog/agentic-ai-benchmarks-guide-what-they-are-how-they-work https://kodemsecurity.com/resources/cve-2026-22778-critical-remote-code-execution-in-vllm-multimodal-inference https://krafton.ai/blog/whale/ https://labs.cloudsecurityalliance.org/agentic/agentic-mcp-security-best-practices-v1 https://labs.cloudsecurityalliance.org/research/csa-research-note-huggingface-autonomous-agent-3-202607 https://labs.cloudsecurityalliance.org/research/csa-research-note-langchain-langgraph-critical-vulns-framewo https://labs.cloudsecurityalliance.org/research/csa-research-note-langchain-langgraph-vulnerabilities-202603 https://labs.scale.com/papers/harnessopt-bench https://lilianweng.github.io/posts/2026-06-24-scaling-laws/ https://lilianweng.github.io/posts/2026-07-04-harness/ https://linjohnss.github.io/scal3r https://lyceum.technology/magazine/vllm-vs-sglang-vs-tensorrt-llm-2026-picking https://mcp.csdn.net/6a2e4df2662f9a54cb7eeb74.html https://mem0.ai/blog/state-of-ai-agent-memory-2026 https://microsoft.com/en-us/research/blog/echoverse-deep-evolving-environments-for-agent-and-use-agents/ https://microsoft.com/en-us/research/blog/orchard-an-open-framework-for-scalable-agent-and-ai/ https://motherduck.com/blog/obsidian-rag-duckdb-motherduck https://nathanbenaich.substack.com/p/e https://nathanbenaich.substack.com/p/europe-ai-sovereignty https://nathanbenaich.substack.com/p/fund-iii https://nathanbenaich.substack.com/p/state-of-ai-may-2026 https://openai.com/index/hugging-face-model-evaluation-security-incident https://opensourcedofu.com/2026/08/truefoundry-launches-trueforge https://opentrain.ai/papers/swe-library-agent-joint-generation-and-maintenance-of-multiple-applications-be--arxiv-2608.29310 https://opentrain.ai/whale-a-simple-agent-and-joint-harness-weight-optimization https://parallel.ai/articles/what-is-an-agent-harness https://particula.tech/blog/sglang-vs-vllm-inference-engine-comparison https://pawbench.github.io https://perplexity.ai/bumblebee https://preprints.org/manuscript/202604.0428/v1 https://previewsworld.com/Catalog/APR260680 https://pulumi.com/blog/building-ai-agents-cli-vs-mcp https://research.google/blog/chain-of-agents-large-language-models-collaborating-on-long-context-tasks https://rice-robotpi-lab.github.io/RoboTok https://shadowdancer-1.github.io https://simonwillison.net/2026/Aug/7/openai-timeline/ https://simonwillison.net/2026/Aug/8/now-we-have-a-timeline-of-the-openai-accidental-attack-against-h https://simonwillison.net/2026/Jul/28/anatomy-of-a-frontier-lab-agent-intrusion https://simonwillison.net/2026/Jul/31/stateless-mcp/ https://simonwillison.net/2026/Sep/24/harder/ https://simonwillison.net/2026/Sep/3/gpt6-astra https://snyk.io/news/snyk-2026-state-of-agentic-ai-adoption-volume-ii https://sparai.org/projects/f26/recPcCQjmdWCEjPJL https://sparai.org/projects/f26/recPcQcjmdWCEjPJL https://sparai.org/projects/f26/recPcQcjmdWCEjPJL https://sparai.org/projects/f26/recPcQcjmdWCEjPJL https://spheron.network/blog/vllm-vs-tensorrt-llm-vs-sglang-benchmarks https://syncsoft.ai/en/blog/agentic-rag-evaluation-metrics-2026 https://sysdig.com/blog/cve-2026-33626-how-attackers-exploited-lmdeploy-llm-inference-engines-in-12-hours https://theaiengineer.substack.com/p/agentic-rag-vs-cua-vs-a2a https://theaiengineer.substack.com/p/the-2026-ai-agent-stack-drawn-from https://theaiengineer.substack.com/p/the-2026-ai-agents-stack-2026-edition https://theaiengineer.substack.com/p/the-agents-stack-2026-edition https://theaiengineer.substack.com/p/the-ai-agents-stack-2026-edition https://theaiengineer.substack.com/p/vllm-vs-ollama-vs-sglang-vs-tensorrt https://theaiengineer.substack.com/p/why-ai-agents-keep-failing-in-production https://thenewstack.io/model-context-protocol-roadmap-2026 https://usewire.io/blog/long-context-vs-rag-what-the-data-shows https://velesdb.com https://vllm.ai/blog/2026-07-22-kimi-k3-preview https://wavect.io/blog/rag-vs-finetune-vs-longcontext-2026 https://witness.ai.com/b/rag-security https://www.amazon.science/publications/keyword-search-is-all-you-need-achieving-rag-level-performance-without-vector-databases-using-agentic-tool-use https://www.braintrust.dev/articles/best-ai-agent-debugging-tools-2026 https://www.cnblogs.com/gpustack/p/19568201 https://www.cncf.io/ https://www.cockroachlabs.com/blog/a2a-agent-state-data-layer/ https://www.databricks.com/blog/big-context-rag-performance-llms https://www.datadoghq.com/state-of-ai-engineering https://www.freecodecamp.org/news/ai-evaluation-engineering-build-and-production-grade-p-llm-evaluation-platform-handbook https://www.langchain.com/state-of-agent-engineering https://www.linkedin.com/posts/claudio-stamile_if-youre-building-agents-youve-probably-activity-7416401402438205440-t9V_ https://www.microsoft.com/en-us/research/blog/echoverse-deep-evolving-environments-for-computer-use-agents/ https://www.microsoft.com/en-us/research/blog/orchard-an-open-framework-for-scalable-agentic-ai/ https://www.opensourceforu.com/2026/08/truefoundry-launches-trueforge https://www.opentrain.ai/papers/remember-when-it-matters-proactive-memory-agent-for-long-horizon-agents--arxiv-2607.08716 https://www.preprints.org/manuscript/202604.0428/v1 https://www.reuters.com/business/anthropic-ceo-urges-ai-companies-slow-model-development-2026-09-12 https://www.sysdig.com/blog/cve-2026-33626-how-attackers-exploited-lmdeploy-llm-inference-engines-in-12-hours https://www.theatlantic.com/technology/2026/09/dario-amodei-slow-down-ai-save-humanity/688610 https://www.upwork.com/blog/introducing-the-upheld-dataset-upwork-human-evaluated-longer-dialogues https://www.vectara.com/blog/context-engineering-and-you-trust-long-context https://x.com/AndrewYNg/status/2098459474608672916 https://x.com/AnthropicAI/status/2098097512544444447 https://x.com/Thom_Wolf/status/2098080470235762702 https://x.com/cwolferesearch https://x.com/emollick/status/2098812026944454715 https://x.com/hwchase17/status/2021671063272796549 https://x.com/karpathy/status/2098811935114551617 https://x.com/omarsar0/status/202762406204076532 https://x.com/sama/status/2097776310940569783 https://xiyuanyang45.github.io/WMRL/ https://zenn.dev/retrieva_tech/articles/f4386f6b246b34 https://zli12321.github.io/LHTB https://github.com/griver/Q-RAG https://github.com/anthropics/claude-sonnet-5.5 https://github.com/anthropics/claude-opus-5.5 https://github.com/bytedance/deer-flow https://github.com/cognition-labs/devin-fusion https://github.com/bytedance/agent-harness-omni-io https://arxiv.org/abs/2607.11433 https://arxiv.org/abs/2607.08964 https://arxiv.org/abs/2609.35427 https://arxiv.org/abs/2609.37559 https://arxiv.org/abs/2609.37673 https://arxiv.org/abs/2609.32722 https://arxiv.org/abs/2609.34392 https://arxiv.org/abs/2609.36322 https://arxiv.org/abs/2609.39982 https://alphaxiv.org/abs/2609.02783 https://alphaxiv.org/abs/2609.16591 https://blog.cloudflare.com/agent-access-model https://labs.cloudsecurityalliance.org/research/csa-research-note-huggingface-autonomous-agent-breach-202607 https://microsoft.com/en-us/research/blog/echoverse-deep-evolving-environments-for-computer-use-agents/ https://microsoft.com/en-us/research/blog/orchard-an-open-framework-for-scalable-agentic-ai/ https://opentrain.ai/papers/super-library-agent-joint-generation-and-maintenance-of-multiple-applications-be--arxiv-2608.29310 https://opentrain.ai/whale-a-simple-recipe-for-joint-harness-weight-optimization https://sparai.org/projects/f26/recPcQcjmdWCEjPJL https://sparai.org/projects/f26/recPcQcjmdWCEjPJL https://witness.ai/blog/rag-security https://www.databricks.com/blog/long-context-rag-performance-llms https://www.freecodecamp.org/news/ai-evaluation-engineering-build-a-production-grade-llm-evaluation-platform-handbook https://www.microsoft.com/en-us/research/blog/echoverse-deep-evolving-environments-for-computer-use-agents/ https://www.microsoft.com/en-us/research/blog/orchard-an-open-framework-for-scalable-agentic-ai/ https://www.vectara.com/blog/context-engineering-can-you-trust-long-context https://x.com/hwchase17/status/2101671063272796549 https://x.com/omarsar0/status/2102762406204076532 https://arxiv.org/abs/2609.34563 https://arxiv.org/abs/2609.24555 https://arxiv.org/abs/2609.38349 https://arxiv.org/abs/2609.39154 https://arxiv.org/abs/2609.38792 https://arxiv.org/abs/2609.39788 https://huggingface.co/blog/Hcompany/holo4 https://orca.security/resources/blog/sglang-llm-framework-rce-vulnerabilities https://www.anthropic.com/news/claude-frontier-academy https://stratechery.com/2026/whos-afraid-of-chinese-models https://huggingface.co/blog/agent-glossary https://huggingface.co/blog/microsoft/thinkingbox https://huggingface.co/blog/icml-2026-open-reproductions https://github.com/kennethlaw325/awesome-llm-knowledge-systems https://www.coddykit.com/pages/blog-detail?id=5129964&slug=the-state-of-ai-agents-october-2026-what-30-days-of-github-trending-reveals-abou https://medium.com/codex/mlops-in-2026-from-mlflow-to-llmops-the-complete-guide-to-shipping-ai-in-production-0024955b70c4 https://thenewstack.io/microsoft-agent-framework-go https://atlan.com/know/ai-agent/microsoft/agent-framework https://www.langchain.com/resources/ai-agent-frameworks https://www.digitalapplied.com/blog/microsoft-agent-framework-1-0-dotnet-python-guide https://tech-insider.org/anthropic-claude-enterprise-frontier-safeguards-2026 https://juliangoldie.com/anthropic-enterprise-frontier-safeguards https://aiweekly.co/ai-news-today/anthropic-news https://blog.mean.ceo/anthropic-claude-news-october-2026 https://arxiv.org/abs/2609.16816 https://www.microsoft.com/en-us/research/blog/agent-lightning-v1-0-a-3500-line-lightweight-agentic-rl-framework-for-training-agents-with-real-harnesses https://arxiv.org/html/2610.12299v1 https://huggingface.co/papers/2610.12299 https://arxiv.org/abs/2610.10533 https://arxiv.org/abs/2610.11794 https://arxiv.org/abs/2610.11169 https://gofastmcp.com/servers/auth https://bigdataboutique.com/blog/building-mcp-servers-with-fastmcp-7-mistakes-to-avoid https://www.giskard.ai/knowledge/cot-forgery-an-llm-vulnerability-in-chain-of-thought-prompting https://www.oneusefulthing.org/p/the-dot-and-the-swarm https://blogs.windows.com/windowsdeveloper/2026/10/07/microsoft-execution-containers-policy-driven-containment-for-ai-agents https://github.com/microsoft/agent-lightning https://aws.amazon.com/bedrock/glm https://www.explainx.ai/news https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-zai-glm-5-3.html https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-zai-glm-5-3.html https://aws-news.com/article/2026-10-06-glm-53-by-zai-is-now-generally-available-on-amazon-bedrock https://superpowerdaily.com/posts/aws-adds-z-ai-s-glm-5-3-to-bedrock-for-eligible-enterprise-customers https://explainx.ai/blog/glm-5-3-amazon-bedrock-enterprise-coding-model-ids-2026 https://x.com/Zai_org/status/2107272032073437185 https://arxiv.org/abs/2610.12341 https://arxiv.org/abs/2610.13735 https://github.com/logly/mureo https://github.com/existence-master/Sentient https://github.com/h100envy/crawl-hive https://github.com/yuk27/VBoyColor https://github.com/Imbad0202/academic-research-skills https://github.com/chekusu/yomado https://github.com/Fei2411/DeepSeek-Bot https://github.com/bherbruck/solidcraft https://github.com/aipoch/open-science https://github.com/mikhaeelatefrizk/affect-labeling-review https://github.com/Maxwooow/Ricoh-GRMod https://github.com/MCG-NJU/VideoJev https://github.com/xerj-org/xerj https://github.com/JustasMonkev/mcp-accessibility-scanner https://github.com/neatlogs/neatroots https://github.com/tkitanishi/PaperBot https://github.com/odakin/arxiv-digest https://github.com/senna-lang/arxiv-compass https://github.com/LiaojunChen/Arxiv-daily https://github.com/semrincaballero/Polymarket-Prediction-Bot https://github.com/arad01486-hash/studio-pulse-relay https://github.com/joiha-steven/quireink https://github.com/YaoGaoChaoLe/EAgent https://github.com/UnderleveledBuilder/awesome-short-drama-skills https://github.com/wozengyi/embodied-daily https://github.com/Aitherium/awresearch https://github.com/KalarisLabs/research-agent-skills https://github.com/xerj-org/xerj https://github.com/metadist/synaplan https://github.com/paulmil11/decks https://github.com/Fangx-AI/geo-doctor https://github.com/C4rohan/Lampline https://github.com/MemTensor/MemT-1.0 https://github.com/smaniches/semantic-scholar-mcp https://github.com/0xward/Thesis-AI### 7.7 CVE、DOI 与 URL(沿用 v112 + v117 新增 0 件 CVE)
v122 沿用 27 件 CVE + 0 件 net-new:CVE-2025-32711 CVE-2025-49596 CVE-2026-14890 CVE-2026-22778 CVE-2026-26029 CVE-2026-3059 CVE-2026-3060 CVE-2026-30623 CVE-2026-3172 CVE-2026-33626 CVE-2026-3989 CVE-2026-4372 CVE-2026-5059 CVE-2026-54769 CVE-2026-57572 CVE-2026-59726 CVE-2026-61447 CVE-2026-73558。v76 6 件 CVE 备料锚入:CVE-2025-68664 + CVE-2026-34070 + CVE-2025-67644 + CVE-2025-64439 + CVE-2025-6984 + CVE-2024-5998。v116 净增 3 件 CVE:CVE-2025-6514 mcp-remote OAuth 命令注入 CVSS 9.6 + CVE-2026-48710 Starlette fakeredis 版本锁定 + CVE-2026-33626-already-counted = 27 件总览。DOI:10.1145/3770854.3780171 DOI:10.1145/3770855.3818074 DOI:10.48550/arXiv.2602.23368 DOI:10.1145/3803798 = 4 件总览。URL v118 沿用 378 件 + v119 新增 8 件 + v120 net-new 6 件 + v121 net-new 3 件 + v122 net-new 8 件 + v123 net-new 7 件 + v124 net-new 8 件 = 418 件总览(完整 URL 列表见 archive/llm-application-changelog.md v119 段).
本次变更
v124 · 2026-10-20
v124 = 10-19 18:00 → 10-20 18:00 ≈ 24h E1 活文档增量(manifest e1_live:llm-application 工作集 8 件 repo_card,无 paper_card / arXiv ID)= v124 net-new 0 件 arXiv unique ID(承接稳态期第 5 段连续无 arXiv)+ 0 件 paper_card + 8 件 repo_card。承接稳态期延续 + v116-v123 已立骨架沿用不动。
v124 核心发现 ⚠⚬:Agent 应用外壳(shell)层正在独立成一等工程品类。8 件材料中 6 件不涉及模型与训练,只解决"把模型接进真实工作流"的最后一公里 —— 与 v116-v123 已立的"训练侧 harness"(HarnessDev / EMHO / Mara Chain / Harness-of-Harness)形成镜像。新增"接入点轴"横切视图,六类接入点单日各有独立开源实现:① 检索(xerj)② 对话/网页(synaplan + geo-doctor)③ 文档协作(decks + Thesis-AI)④ 语音/电信(Lampline)⑤ 学术检索(semantic-scholar-mcp)⑥ 记忆基础设施(MemT-1.0)。新增外壳层三命题 ⚠⚬:命题 A 本地优先是默认架构(与 rogue agent / HF breach / MCP 无认证安全事件形成因果闭环);命题 B 第一 KPI 是成本可见性而非能力(与 AgentSysBench 成本中心结论同向);命题 C 外壳层正绕过 MCP 生态另起一套分发形态。
v124 工作集 8 件处理清单:① xerj-org/xerj(10669★ +12131/周 Apache-2.0 Rust · production · 单 Rust 二进制索引 code/docs/logs/JSONL/CSV/PDF · BM25+向量+混合 · 免 grep 代码搜索 · agent memory 与 RAG · Elasticsearch 兼容 API · Corpus Hub 数百参考记忆语料)→ v120 主轴 anchor 沿用稳态第 3 例且发生实质范围扩张 ⚠⚬⚬⚠(形态扩为"搜索引擎 + MCP server"、语料扩至 logs/JSONL/CSV/PDF、Corpus Hub 竞争 Memory 生态位),单周 +12131 ≈ 存量 1.4 倍为本轮唯一真实增长信号,也是 v120 "40× token 节省 vs grep"最完整溯源;② metadist/synaplan(201★ Apache-2.0 PHP/Symfony experimental · 自托管 AI 平台 · 任意模型聊天 + 自有文档 RAG + 可嵌入 widget + MCP · 逐条回答成本可见)→ 对话/网页接入栖预备第 1 例 ⚠⚬,构成去 Python 化第 3 例证据;③ paulmil11/decks(85★ MIT HTML experimental · 幻灯片下留言 → Claude Code/Codex 改 → 在该幻灯片回帖)→ 文档协作接入栖预备第 1 例 ⚠⚬,第 3 件"在既有工件上留言、由 coding agent 回改"交互范式件;④ Fangx-AI/geo-doctor(90★ MIT JavaScript experimental · 待LLM分类 · 中英双语 GEO skill · 查清豆包/DeepSeek/ChatGPT/Claude 为何不推荐你 → 改资料 → 复查)→ 增长/归因接入栖预备第 1 例,partial anchor(无任何量化指标)⚠⚬;⑤ C4rohan/Lampline(118★ Apache-2.0 Python experimental · 导入 Dialogflow CX / CX Agent Studio agent + 自有硬件经 jambonz 接听电话 + 20 家模型供应商)→ 语音/电信接入栖预备第 1 例 ⚠⚬,本轮最重的存量系统迁移样本;⑥ MemTensor/MemT-1.0(98★ Python · NOASSERTION · 待LLM分类 · 官方仓库)→ partial anchor(card 无技术描述/参数/评测)⚠⚬,但归属 Memory 主线官方仓库,与 xerj Corpus Hub 分处"自建记忆栈 vs 可复用记忆语料分发"两极;⑦ smaniches/semantic-scholar-mcp(23★ MIT Python experimental · 14 个类型化工具(论文检索/引用图谱/作者画像/推荐)· PyPI+Docker · 学术档案完整 · 相关度 95 deterministic)→ academic discovery 栖 v124 第三例 anchor ⚠⚬,补出 RSS 聚合/主题跟踪/具身专项之外的第四型:类型化 API 封装;⑧ 0xward/Thesis-AI(2★ MIT TypeScript experimental · 给定知识库 → 规范学术散文 · 学术档案完整 · 相关度 80 deterministic)→ academic-writing 栖 v124 第四例 anchor,规模不足不升档 ⚠⚬,做"受控输入 vs 自主取证"分野(试金石"证据链可审计"分界线)。
v124 方法学警示 ⚠⚬:本轮 8 件全部周增 +0、5 件 2026-10-20 首采 —— 周增全零且多为首次采集 = GitHub Trending 采样窗口噪声日信号,不能据此判断热度衰减;唯一真实增长为 xerj(+12131/周)。
v124 趋势新增:⑧ 外壳层独立成一等工程品类 ⚠⚬;⑨ 外壳层与 Memory 生态位开始重叠 ⚠⚬⚬ —— xerj 宣称 "agent memory and RAG" 且 Corpus Hub 提供数百参考记忆语料,与 Memory 第十六栖(Mem0/MemOS/Cognee/MemSearch)、第十七栖(通信开销型压缩)、第十八栖(具身文献发现)竞争同一生态位,本轮最值得后续 3-5 棒跟踪的张力 ⚠⚬⚬。
v124 新增开放问题 Q124.571-Q124.577 7 项(外壳层接入点轴是否独立成类 / xerj +12131 stars 成因归因与 Corpus Hub 生态位竞争 / BM25 四合一收益口径不可与向量库直接相减 ⚠⚬ / 外壳层与 MCP 竞合边界 / GEO skill 可核查判定标准缺失 ⚠⚬ / Lampline 迁移 Dialogflow CX 保真度 / 学术工具池七件尚无可横向比较效果学证据不升档)。
v124 net-new 数量:arXiv 1334 + 0 = 1334;CVE 27 + 0 = 27;DOI 4 + 0 = 4;URL 410 + 8 = 418;paper_card 1735 + 0 = 1735。反思棒第 89 例 + 监控第 95 次。
v124 · 2026-10-20 · Wave3 E1 第五棒(承接稳态期第 5 段连续无 arXiv/paper_card 净增 · 8 件 repo_card)· + 外壳层"接入点轴"第 1 例 ⚠⚬ + 外壳层三命题 ⚠⚬ + 去 Python 化第 3 例证据 ⚠⚬ + v120 xerj 主轴 anchor 第 3 例且范围扩张 ⚠⚬⚬⚠ + Memory 生态位竞争张力 ⚠⚬⚬ + 2 件 partial anchor(geo-doctor + MemT-1.0);arXiv 1334 / CVE 27 / DOI 4 / URL 418 / paper_card 1735**
v123 · 2026-10-19
v123 = 10-18 18:00 → 10-19 18:00 ≈ 24h E1 活文档增量(manifest e1_live:llm-application 工作集 8 件 repo_card,无 paper_card / arXiv ID)= v123 net-new 0 件 arXiv unique ID(承接稳态期延续)+ 0 件 paper_card + 8 件 repo_card(4 主轴 anchor + 3 边沿 partial + 1 partial)。承接稳态期延续 + v117-v122 已立骨架沿用不动 + 立标池回稳期第 19 日承接稳态 ⚠⚬⚬ + HF Daily 10-19(Self-Evolving Agent Harness 202▲ #5 + MAI-Gym 155▲ + Mosaic 54▲ + OmniVerifier-C1 32▲ + Each Reward to Its Relevant Tokens 56▲)+ frontier lab 公告密度 10-13→10-19 7 日延伸稳态 ⚠⚬⚬⚠ 沿用 + Agent 框架生态成熟化拐点 + Harness 设计自动化四路径 + Hebero vs RobotWorld + frontier lab 商业模式三栖叙事 + Memory 第十六/十七栖 + Mem0 LoCoMo benchmark + AgentSysBench 沿用稳态 ⚠⚬⚬⚬。
v123 工作集 8 件处理清单:① xerj-org/xerj(8618★ +2401 Apache-2.0 Rust · autoindex · 40× token 节省 vs grep · Elasticsearch 兼容 · x10 基础设施成本节省)→ v120 主轴 anchor 沿用稳态 第 2 例 ⚠⚬⚬⚬ + Memory 第十七栖"通信开销型 Memory 压缩"栖预备第 1 例沿用稳态 ⚠⚬⚬ + 5902★ → 8618★ trending 升级;② arad01486-hash/studio-pulse-relay(72★ HTML experimental · Roblox Studio Live Sync 2026 · Open Cloud WebSocket Agent · Playtest Hot-Reloading)→ Roblox 真人游戏开发测试热重载 CUA 边沿栖预备第 1 例 ⚠⚬,partial anchor 周增 0 提示"已建立但停滞";③ joiha-steven/quireink(37★ +3 NOASSERTION TypeScript experimental · 自托管博客 AI 代为运行 · 一个 Bun 进程两个 SQLite · 无云端账号)→ local-first 自托管博客 AI 代运栖预备第 1 例 ⚠⚬;④ YaoGaoChaoLe/EAgent(103★ GPL-3.0 HTML experimental · 0 token / 0 门槛 · DeepSeek 网页版 Agent · DeepSeek 操控电脑)→ CUA 轻量化接驳 DeepSeek 浏览器栖预备第 1 例 ⚠⚬ + 中国国内 SaaS LLM→ 本地执行的"无 LLM API key、无 token 计费"实验;⑤ UnderleveledBuilder/awesome-short-drama-skills(71★ CC0-1.0 HTML experimental · AI 短剧/漫剧 Skill 精选 · 剧本/改编/分镜 · Seedance / Kling / MiniMax 视频提示词 · Claude Code / Codex / OpenClaw 通用)→ multimodal 主轴短剧/分镜 Skill 链栖预备第 1 例 ⚠⚬⚬ + 跨 agent harness Skill 复用实证 · 与 Skill Constellations arXiv:2610.11169 承接稳态;⑥ wozengyi/embodied-daily(JavaScript experimental · 具身 Daily · HF Daily Papers + arXiv 联合每日具身 AI 论文推荐 · acceptance anchor)→ Memory 第十八栖"具身文献发现"栖预备第 1 例 anchor ⚠⚬⚬⚬ + academic discovery 栖 v121 双 anchor → v123 三 anchor(senna/arxiv-compass + LiaojunChen/Arxiv-daily + embodied-daily)扩位 ⚠⚬⚬;⑦ Aitherium/awresearch(Apache-2.0 Python production · 提问研究问题 → 可核验的带引文报告 · academic-writing tool)→ RAG + academic-writing 栖预备第 1 例 ⚠⚬,partial anchor 与 v119 Imbad0202/academic-research-skills + aipoch/open-science + v121 SakanaAI/AI-Scientist-v3 + v123 KalarisLabs 形成"academic-writing tool 池"实证扩位;⑧ KalarisLabs/research-agent-skills(43★ NOASSERTION Python experimental · Research Agent Skills 学术研究 Skill 集 · AI/ML/生物/化学/医学/物理 · 学术写作/文献综述/引文/科学工作流)→ 学术写作栖预备实证第 2 例 anchor ⚠⚬ + 学术研究 Skill 体系双栖预备(Imbad0202 ↔ KalarisLabs)⚠⚬ + Skill Constellations arXiv:2610.11169 子集实例 ⚠⚬。
v123 net-new 数量:arXiv 1334;URL 403 + 7 = 410;paper_card 1735。反思棒第 88 例 + 监控第 94 次。
v123 · 2026-10-19 · Wave3 E1 第四棒(承接稳态期第 4 段累积日无 arXiv/paper_card 净增 · 8 件 repo_card)· + Memory 第十八栖"具身文献发现"栖预备第 1 例 anchor ⚠⚬⚬⚬ + academic discovery 三 anchor 扩位 ⚠⚬⚬ + v120 xerj 主轴 anchor 沿用 第 2 例 ⚠⚬⚬⚬ + multimodal 主轴短剧/分镜 Skill 链栖预备 第 1 例 ⚠⚬⚬ + 学术研究 Skill 体系双栖预备 ⚠⚬ + CUA 轻量化接驳 DeepSeek 浏览器栖预备 第 1 例 ⚠⚬ + local-first 自托管博客 AI 代运栖预备 第 1 例 ⚠⚬ + RAG + academic-writing 栖预备 第 1 例 ⚠⚬ + Roblox Live Sync 2026 边沿 CUA 栖预备 第 1 例 ⚠⚬;arXiv 1334 / CVE 27 / DOI 4 / URL 410 / paper_card 1735**
v122 · 2026-10-18
v122 = 10-17 18:00 → 10-18 18:00 ≈ 24h E1 活文档增量(工作集 8 件 paper_card)= v122 net-new 3 件 arXiv + 8 件 paper_card + 8 件 arXiv URL。八件工作集 4+2+2 三栖:Forms-of-LLM-Integrated-Applications + Agent Plasticity + Inherit-MAS + Self-Retrospection Distillation 主栖;Geometry of Hierarchical Navigation + Is Memorization Context-Sensitive 邻接栖;ORCAGen + OmniCapBench 边沿栖。v122 数量:arXiv 1331 + 3 = 1334;URL 395 + 8 = 403;paper_card 1727 + 8 = 1735。
v122 · 2026-10-18 · Wave3 E1 第三棒 · 沿用稳态全栈承接 v117-v121 立标池回稳期;arXiv 1334 / URL 403 / paper_card 1735**