论文

Agent Lightning v1.0 是一个轻量级的可控 Agentic RL 框架，约 3500 行代码实现，支持任意 Agent harness，并作为研究 retokenization、样本合并、优势计算、损失归一化与后端调度等挑战的实用测试平台。Agent Lightning v1.0 is presented, a lightweight framework for harnessed agentic RL implemented in approximately 3,500 lines of code that supports arbitrary agent harnesses and serves as a practical testbed for studying challenges in retokenization, sample merging, advantage calculation, loss normalization, and backend scheduling.

本文设计了一项对比研究，结合受控定量实验与配对轨迹分析，并将观察结果归纳为一个包含三个高层类别与十二种 Skill 使用模式的分类法，表明当噪声轨迹转化为稳定执行的过程性锚点时，Skill 便会发挥作用。This work designs a contrastive study that combines controlled quantitative experiments with paired trajectory analysis and consolidates observations into a taxonomy of three high-level categories and twelve skill-use modes, showing that skills work when noisy trajectories become procedural anchors that stabilize execution.

研究表明矛盾解析本质上是写入时并发控制，并将缺失的契约——一个在隔离性、模式与来源维度上被证明正确的写入时正确性规范——显式化，固定了每个生产启发式都默认假设、却没有任何已部署系统显式给出的保证。It is shown that contradiction resolution is write-time concurrency control and make the missing contract explicit, a write-time correctness specification, proved sound across isolation, schema, and provenance, pinning the guarantee every production heuristic assumes but no deployed system makes explicit.

本文提出了 LIFE-RL 框架，在不修改内部控制流的前提下，将原生编码 Agent harness 与可扩展的策略梯度优化相连接，并通过 GSPO 在三个原生编码 Agent harness 上训练稀疏 MoE 模型 Qwen3.5-35B-A3B 对其进行了评估。LIFE-RL, a framework that bridges native coding-agent harnesses with scalable policy-gradient optimization without modifying their internal control flow, is presented and evaluated by training the sparse MoE model Qwen3.5-35B-A3B with GSPO across three native coding-agent harnesses.

本文提出 Co-RL，一种由多个解耦模型组成的框架，这些模型不共享参数，通过基于彼此输出奖励的强化学习同时进行优化，并表明无监督推理可以通过协作式多智能体训练涌现This work introduces Co-RL, a framework in which multiple decoupled models, sharing no parameters, are simultaneously optimized through RL using rewards derived from their peers, and shows that unsupervised reasoning can emerge through cooperative multi-agent training.

论文认为，agentic CTI 调查的瓶颈在于该 substrate 而非模型能力，并提出了面向 Agent 的语料库脚手架 CTIFoundry。It is argued that this substrate, not model capability, is the bottleneck on agentic CTI investigation, and CTIFoundry, an agent-native corpus scaffold, is presented.

SkillGate 将 9B 策略的成功率从 40.8% 提升至 53.2%，显著优于将相同预算仅用于 outcome reward 的方案，同时将误导性候选的暴露减少三分之二，并读取更少的 skill。SkillGate lifts a 9B policy from 40.8% to 53.2% trial success, well ahead of the identical budget spent on outcome reward alone, while cutting exposure to misleading candidates by two thirds and reading fewer skills.

受损模型评估测试在第一个合法 tool call 之后插入 ground-truth 攻击调用，从而独立于模型行为测试 APC，证明了 APC 实现的 Blast Radius 单调性与组合可靠性。The compromised-model evaluation tests APC independently of model behavior by inserting the ground-truth attack call after the first legitimate tool call, which proves Blast Radius Monotonicity and Composition Soundness for APC implementations and proves Blast Radius Monotonicity and Composition Soundness for APC implementations.

提出 Repo0，一个面向零到全代码生成的持续结构演化框架，维护显式的架构状态，实例化为双有向无环图（Dual-DAG），由需求级 DAG、组件级 DAG 及其对齐关系组成。Repo0 is presented, a continuous structural evolution framework for zero-to-all code generation that maintains an explicit architectural state instantiated as a Dual-Directed-Acyclic-Graph (Dual-DAG), consisting of a requirement-level DAG, a component-level DAG, and their alignment relation.

一项配对消融实验在保留仓库与可执行工程上下文的同时移除显式科学指导，表明科学知识并非一律有益：可靠信息能约束修复、提升平均表现与 token 效率，而错位指导则会诱发锚定，不必然提升精确修复成功率。A paired ablation that removes explicit scientific guidance while preserving the repository and executable engineering context shows that scientific knowledge is not uniformly beneficial: well-grounded information can constrain repair and improve average performance and token efficiency, whereas poorly aligned guidance can induce anchoring and does not necessarily improve exact repair success.

本文指出，持续技能演化的关键瓶颈既非编辑能力亦非迭代轮数，而在于评估反馈能否持续提供可信的演化梯度；据此提出 SkillEvo，由可信反馈生成梯度，由可控治理约束方向。This work argues that the binding constraint on sustained skill evolution is neither editing capability nor the number of iterations, but whether the evaluation feedback keeps supplying trustworthy evolution gradients, and introduces SkillEvo, in which trustworthy feedback generates the gradient and controllable governance constrains its direction.

泄露带来两类现实攻击：一个训练好的分类器可从常规自然语言输出中推断用户记忆的语义谓词；一个由 RL 训练的对抗者可从生产级风格的 Agent 中完整提取社会安全号码。Leakage enables two practical attacks: a trained classifier that infers semantic predicates about user memories from routine natural-language outputs, and an RL-trained adversary that extracts full Social Security Numbers from a production-style agent.

Bespoke-Card 在传统通用估计器与学习型估计器架构之外开辟了一条新的基数估计路径，它是一个 Agent 驱动的系统，将面向特定工作负载的基数估计器合成为可执行代码。Bespoke-Card is opening a new avenue for cardinality estimation next to classical generic estimators and learned estimator architectures, an agent-driven system that synthesizes workload-specific cardinality estimators as executable code.

大语言模型 Agent 可在推理时构建工作流以适配复杂任务，但单个回合中发现的过程通常在执行后即被丢弃。现有 Skill 库提供可复用的可执行例程，但通常离线组装，且无法从 Agent 自身的工作流中不断生长。我们提出 FlowEvo，一个无需训练、推理时工作流与 Skill 协同演化的框架。FlowEvo 将成功的工作流编译为可调用 Skill，存入持久化库，并通过直接执行或作为上下文的方式使用检索到的 Skill。Large language model agents can adapt to complex tasks by constructing workflows at inference time, but procedures discovered in one episode are usually discarded after execution. Existing skill libraries provide reusable executable routines, but are typically assembled offline and do not grow from the agent's own workflows. We introduce FlowEvo, a training-free framework in which workflows and skills co-evolve at inference time. FlowEvo compiles successful workflows into callable skills, stores them in a persistent bank, and uses retrieved skills either through direct execution or as context

LLM 编程 Agent 通过接口下发 Bash 命令，这些接口可能对模型输出进行序列化、包装与再解析。仅凭匹配的执行得分无法区分命令生成阶段的错误与生成后引入的失败。QuoteBench 围绕一道刻意未转义的额外解析器，将生成合约与执行传输交叉，在 14 个源自真实事件的家族、共 56 个一次性任务上以精确最终状态校验来度量该边界。在插值点进行转义可复现每条回放回复的原始路径结果，因此任何在已声明边界下取得的恢复成效都必须……LLM coding agents issue Bash commands through interfaces that may serialize, wrap, and reparse model output. Matched execution scores alone cannot distinguish command-generation errors from failures introduced after generation. QuoteBench measures this boundary with exact final-state validation on 56 one-shot tasks from 14 incident-derived families, crossing the generation contract with the execution transport around one deliberately unescaped added parser. Escaping at the interpolation point reproduces each replayed reply's raw-path outcome, so any recovery under a disclosed boundary must com

使用工具的 AI agents 将 delegated 任务转化为 provider effects，但 authorization 常止于准入，而 provider 状态、delivery、retry 与 recovery 在持续演化。请求可能在 commit 前变化，或响应丢失使替换从同一 approval 产生二次 effect。我们提出 AID-Guard，一种有状态的 authorization-to-effect 闭合协议：commit 时重新验证已批准请求与 provider 状态，模糊情形下仅保留一个 reservation，并在出现终结结果或经 delivery fence 认证的无 effect 后才允许 release 或一次后继动作。对于支持的 pro[vider]……Tool-using AI agents turn delegated tasks into provider effects, yet authorization often ends at admission while provider state, delivery, retry, and recovery evolve. A request may change before commit, or response loss may cause a replacement to create a second effect from one approval. We present AID-Guard, a stateful authorization-to-effect closure protocol. It revalidates the approved request and provider state at commit, retains one reservation under ambiguity, and permits release or one successor only after a terminal result or certified no effect with a delivery fence. For supported pro

本文以 Oracle-to-PostgreSQL 迁移为受控软件转换任务，研究跨 agent 的 specification portability。研究包含两个阶段：首先在 1,006 个 PL/SQL 文件上评估 specification-first 迁移 pipeline，其中 623 个成功重新生成，380 个生成脚本在 PostgreSQL 16 中成功执行；其次使用 Amazon Kiro、Google Gemini 与 GitHub Copilot（含 Claude Code 与 Cursor）在 1,802 个 Oracle 脚本及其对应 PostgreSQL 实现的[数据集]上进行跨 agent 实验。This paper investigates cross-agent specification portability using Oracle-to-PostgreSQL migration as a controlled software transformation task. The study combines two experimental stages. First, a specification-first migration pipeline was evaluated on 1,006 PL/SQL files, of which 623 were successfully regenerated and 380 generated scripts executed successfully in PostgreSQL 16. Second, cross-agent experiments were conducted on a dataset of 1,802 Oracle scripts with corresponding PostgreSQL implementations using Amazon Kiro, Google Gemini, and GitHub Copilot, with Claude Code and Cursor inclu

我们提出 PhysCaP，一种面向机器人操作主动感知的物理信息驱动 code-as-policy agent。尽管 vision-language-action 策略擅长模仿示教，但它们依赖被动观察，无法推断对操作至关重要的潜在物理属性。PhysCaP 在 code-as-policy 框架基础上引入物理信息驱动的探索层，通过交互实现显式信息获取。它提出了无需训练的物理属性提取模块，仅依靠机器人本体感知即可估算物体质量和刚度，无需额外传感器。为了在 balanWe present PhysCaP, a Physics-Informed Code-as-Policy agent for active perception in robotic manipulation. While vision-language-action policies excel at imitating demonstrations, they rely on passive observation and fail to infer latent physical properties critical for manipulation. PhysCaP augments code-as-policy frameworks with a physics-informed exploration layer that enables explicit information-seeking through interaction. It introduces training-free physical property extraction modules that estimate object mass and stiffness from robot proprioception without additional sensors. To balan

大语言模型（LLM）agent 的群体行为无法用单 agent 基准刻画。我们提出 PV-SST，一个基于同行投票的社交平台测试平台，并报告一项独立冻结、预先注册的等曝光实验，涵盖四个话题、四个未使用种子、四个开源权重模型家族以及三个预设的更大模型变体。该实验包含 448 次试验和 112 个完整的"模型 × 话题 × 种子"区组。相对于仅话题对照条件，由同行生成的点赞排序的上一轮同行帖子信息流，会提升最终轮在两端的词汇相似度Population-level behavior in large-language-model (LLM) agents cannot be characterized by single-agent benchmarks. We introduce PV-SST, a peer-voted social-platform testbed, and report a separately frozen, preregistered matched-exposure experiment spanning four topics, four unused seeds, four open-weight model families, and three prespecified larger variants. The experiment comprises 448 trials and 112 complete model-by-topic-by-seed blocks. Relative to a topic-only control, a feed of previous-round peer posts ranked by peer-generated likes increases final-round lexical similarity in both the

近期大语言模型（LLM）已能作为编码 Agent，根据自然语言请求构建完整游戏。游戏开发尤为严苛，因为程序逻辑、视觉与音频内容、界面、交互和可玩性必须在同一可执行制品中协同工作。因此衡量该能力需要同时评测游戏产品与开发过程。现有基准通常通过评估最终制品或孤立的开发阶段来评测 LLM 的游戏开发能力。我们对完整人机协作开发过程的分析Recent large language models (LLMs) can operate as coding agents that build complete games from natural language requests. Game development is especially demanding because program logic, visual and audio content, interfaces, interaction and playability must function together in one executable artifact. Measuring this capability therefore requires evaluation of both game product and the development process. Existing benchmarks often assess the game development capabilities of LLMs by evaluating the final artifact or an isolated development stage. Our analysis of complete human-agent development

开放式真实交互允许多种有效行为：Agent 可直接回答、请求澄清、提供进度更新，或在执行前进行确认。这种灵活性打破了基于分组的 RL 的核心假设：同一分组内对比的 rollout 不再保证行为可比。因此奖励模型对交互风格的偏好可能扭曲相对优势，使优化偏向奖励偏好的行为而非情境适配的行为。我们将其形式化为奖励公平性问题，并提出 ARC（Advantage RegularizatiOpen-ended real-world interaction admits multiple valid behaviors: an agent may answer directly, ask for clarification, provide progress updates, or confirm before acting. This flexibility breaks a core assumption behind group-based RL: rollouts compared within a group are no longer guaranteed to be behaviorally comparable. As a result, reward-model preferences over interaction style can distort relative advantages and steer optimization toward reward-preferred behaviors rather than context-appropriate ones. We formalize this as a reward fairness problem and propose ARC (Advantage Regularizati

一条安全规则与一段情景日志在同一 AI Agent 上下文中争夺 token。当预算溢出时，二者以相同速率被压缩；但只有规则需要精确措辞才能保持可执行性。在 20 种生产环境 Agent 配置下，Claude Code 基于 Sonnet 4.6 的 /compact 提示在一轮压缩后保留 53% 的安全规则，五轮后仅保留 10%。我们将此现象命名为"压缩悬崖"（Compaction Cliff）。我们提出 Knowledge Triage 框架，通过对 Agent 知识库的每一行按类型分类，并为每类配置独立的保留策略来解决该问题。三种确定性 operatA safety rule and an episodic log compete for the same tokens in an AI agent's context. When the budget overflows, both are summarized at the same rate; only the rule needs exact wording to remain enforceable. On 20 production agent configurations, Claude Code's /compact prompt on Sonnet 4.6 preserves 53\% of safety rules after one compaction round and 10\% after five. We name this the Compaction Cliff. We address it with Knowledge Triage, a framework that classifies each line of an agent's knowledge base by type and routes each type through its own retention policy. Three deterministic operat

自主研究系统执行长研究工作流的能力日益增强，但仅靠自动化并不能确保所得流程保持科学严谨性。我们提出 AutoResearch，一个两阶段系统，将 Idea Generation 与 Idea Execution 相连，以同时解决研究想法如何形成以及如何通过实验可靠建立的问题。在 Idea Generation 中，AutoResearch 持续整合新出现的研究信号与累积的领域知识，识别可迁移的机理洞察，并采用多模型生成与跨模型Autonomous research systems are increasingly capable of executing long research workflows, yet automation alone does not ensure that the resulting process remains scientifically grounded. We introduce AutoResearch, a two-stage system that connects Idea Generation with Idea Execution to address both how research ideas are formed and how they are reliably established through experimentation. In Idea Generation, AutoResearch continuously integrates emerging research signals with accumulated domain knowledge, identifies transferable mechanistic insights, and uses multi-model generation and cross-r

本工作提出 QBugLM，一个多 Agent 框架，可自动化量子软件调试流水线，覆盖基于分类法的缺陷注入、基于 LLM 的检测与修复，直至基于仿真的验证，框架无关地支持 OpenQASM 3.0 程序。This work proposes QBugLM, a multi-agent framework that automates the quantum software debugging pipeline, from taxonomy-driven bug injection to LLM-based detection and repair, and finally to simulation-based validation, for framework-agnostic OpenQASM 3.0 programs.

PLENA 是一个软硬件协同设计的系统，采用三条核心优化路径，具备新颖的扁平化 systolic-array 架构以及支持非对称量化方案的高效计算与存储单元（路径 2）。PLENA is a hardware-software codesigned system that applies three core optimization pathways that features a novel flattened systolic-array architecture and efficient compute and memory units that support an asymmetric quantization scheme (Pathway 2).

GLM-5 在真实编码任务中展现出前所未有的能力，在端到端软件工程挑战的处理上超越既有基线，并提出了新颖的异步 Agent RL 算法，进一步提升了 RL 质量。GLM-5 demonstrates unprecedented capability in real-world coding tasks, surpassing previous baselines in handling end-to-end software engineering challenges and proposing novel asynchronous agent RL algorithms that further improve RL quality.

本教程综合了推动这些汇聚方向的算法、系统与设计原则，为数据科学与数据挖掘研究者提供统一视角，涵盖将 LLM、图数据管理、图挖掘、图 ML 与 agentic 计算融合到下一代 graph-native AI 系统中。This tutorial synthesizes the algorithms, systems, and design principles driving these converging directions, offering data science and data mining researchers a unified perspective on integrating LLMs, graph data management, graph mining, graph ML, and agentic computation into next-generation graph-native AI systems.

本文提出一种分层隔离架构，结合策略感知的 ingestion、retrieval-time gating 与共享推理，并通过服务端 agentic 编排加以执行，在为多租户隔离提供天然强制点的同时，允许客户端框架保留对 agent 组合与延迟敏感操作的控制权。A layered isolation architecture combining policy-aware ingestion, retrieval-time gating, and shared inference, enforced through server-side agentic orchestration is introduced, creating natural enforcement points for multitenant isolation while allowing client-side frameworks to retain control over agent composition and latency-sensitive operations.

本文认为，AI agent——即以大语言模型作为主要推理引擎、动态生成与丢弃代码作为工具性资源的系统——的出现构成了对"软件"本身的根本性重构，而非渐进式的工具改进。This paper argues that the emergence of AI agents -- systems where large language models serve as the primary reasoning engine, dynamically generating and discarding code as an instrumental resource -- constitutes a fundamental restructuring of what software is, not an incremental tool improvement.

本文提出 Memanto，一种面向 agentic 人工智能的通用记忆层，挑战了"必须依赖知识图谱复杂度才能实现高保真 agent 记忆"的普遍假设，并取得 SOTA 准确率。Memanto is introduced, a universal memory layer for agentic artificial intelligence that challenges the prevailing assumption that knowledge graph complexity is necessary to achieve high fidelity agent memory and achieves state of the art accuracy scores.

本文探讨 multi-agent system，并指出当前尚未被充分解决的问题，同时探索了 multi-agent system 在区块链系统中的潜在应用，为其在真实分布式系统中的未来发展与落地提供启示。This paper explores multi-agent systems and identifies challenges that remain inadequately addressed, and explores potential applications of multi-agent systems in blockchain systems to shed light on their future development and application in real-world distributed systems.

本文系统梳理了基于 LLM 的现代智能体中记忆的设计、实现与评估方法，覆盖 2022 年至 2026 年初的相关工作，并将 Agent 记忆形式化为一个涵盖时间范围、表示基底与控制策略的三维分类体系。This survey offers a structured account of how memory is designed, implemented, and evaluated in modern LLM-based agents, covering work from 2022 through early 2026, and formalizes agent memory as a three-dimensional taxonomy spanning temporal scope, representational substrate, and control policy.

本文提出 AgenticRAGTracer，这是首个主要由大语言模型自动构建、专为支持逐步验证而设计的 Agentic RAG 基准。AgenticRAGTracer is introduced, the first Agentic RAG benchmark that is primarily constructed automatically by large language models and designed to support step-by-step validation, and is primarily constructed automatically by large language models and designed to support step-by-step validation.

Continuum，一个通过为 KV cache 保留引入 TTL 机制来优化多轮 Agent 工作负载任务完成时间的服务系统，能保持多轮连续性，并降低 Agent 工作流的延迟。Continuum, a serving system to optimize job completion time for multi-turn agent workloads by introducing time-to-live mechanism for KV cache retention, preserves multi-turn continuity, and reduces delay for agentic workflows.

本文对一个自 2026 年 3 月起持续运行的个人助理 Agent 运行时中的静默失败进行纵向研究，该系统包含约 40 个定时任务、8 个 LLM 提供商、一个工具治理代理以及一个知识库记忆层，由 4,286 个单元测试和 827 项治理检查守护。A longitudinal study of silent failures in a personal-assistant agent runtime in continuous production since March 2026, with roughly 40 scheduled jobs, 8 LLM providers, a tool-governance proxy, and a knowledge-base memory plane, defended by 4,286 unit tests and 827 governance checks is presented.

结果表明，纯视觉设置会降低准确率并增加 token 成本，因为 Agent 缺乏足够的符号化细节，需通过重复的视觉查询进行补偿；研究指向一种面向下一代编码 Agent 的实用文本与视觉混合设计。The results show that a strictly vision-only setup degrades accuracy and increases token cost, because agents lack sufficient symbolic detail and compensate with repeated visual queries, and point to a practical hybrid text-and-vision design for next-generation coding agents.