1
Show HN: Sillage a 4 MB memory that lets a frozen language>model remember
用4MB记忆让冻结的大模型记住新知识，极致压缩且效果惊艳。
Article URL: https://github.com/riscoss63/sillage Comments URL: https://news.ycombinator.com/item?id=49439609 Points: 1 # Comments: 0
2
Show HN: I missed the moving blocks, so I built a real Linux disk defragmenter
老Linux用户为情怀打造ext4磁盘碎片整理器，向Windows时代moving blocks美学致敬，GitHub已开源！
Article URL: https://github.com/gbin/defragger Comments URL: https://news.ycombinator.com/item?id=49438865 Points: 2 # Comments: 0
3
谷歌杀入法律 AI 赛道，推出专用 Gemini 工具
谷歌携Gemini下场法律AI赛道，巨头加码垂直应用，律师行业智能化拐点将至
IT之家 8 月 26 日消息，Alphabet 旗下谷歌周二扩展了其 Gemini Enterprise 人工智能平台，推出面向律师和律所的新工具，进一步加剧了科技公司争夺法律行业 AI 需求的竞争。 谷歌表示，Gemini Enterprise for Legal 将帮助律所利用 AI 处理日常…
4
An AI pentest agent that structurally can't hallucinate a vulnerability — and runs offline
用证据门控从结构上杜绝漏洞幻觉的AI渗透测试智能体，全离线运行，安全报告终于不是AI编的故事了
If you point any LLM at a target and ask it to "write a security report," it will confidently invent findings that aren't there: an imagined TLS weakn…
5
Training Proactive and Personalized LLM Agents
让大模型从“被动应答”走向“主动服务”，教你训练会察言观色的个性化AI代理，附完整方法框架。
arXiv:2511.02208v2 Announce Type: replace Abstract: Despite rapid progress, current AI agents are primarily optimized for isolated task completion. We…
6
Forgotten in Weights, Recovered by Tools: Agentic Tool Unlearning for LLM Agents
揭秘大模型智能体如何通过工具实现“遗忘”，为安全与能力平衡提供新思路。
arXiv:2608.21544v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly deployed as tool-augmented agents, where responses can…
7
Serverless BigQuery MCP Agent with Gemini" published: true tags: googlecloud, gemini, ai, python
构建无服务器BigQuery MCP智能体，用Gemini驱动数据决策，附实战示例。
Demystifying Data Decisions: Building a Serverless BigQuery MCP Agent with Gemini & Google ADK title: "Demystifying Data Decisions: Building a Ser…
8
Physical Agentic AI: An Architecture for Orchestrating a Robot Crew with LLMs
用LLM统一调度多台机器人协作，开启物理世界智能体的全新架构思路。
arXiv:2608.22657v1 Announce Type: cross Abstract: Agentic AI frameworks interpret open-ended task goals and decompose them into multi-step plans. Rich…
9
Agentic Security: A Systematization of Tools, Failure Modes, and Design Laws for LLM-Driven Penetration Testing
系统化梳理LLM渗透测试工具链与失败模式，提出关键设计法则，安全研究者必读。
arXiv:2608.21423v1 Announce Type: cross Abstract: Agentic security uses large-language-model (LLM) agents to plan, dispatch, and interpret security to…
10
LLM-based Agents for Forecasting and Prediction: Methods, Training, Evaluation, and Applications
从方法到应用，全面解析LLM智能体如何革新预测科学，一文掌握前沿进展。
arXiv:2608.23058v1 Announce Type: new Abstract: Large language models (LLMs) now support forecasting systems that combine language-based reasoning wit…
11
MemGuard: Persisting Verifier Signals for LLM-Agent Memory Governance
LLM智能体长任务中记忆失效的痛点，MemGuard用验证信号让存储经验长期可靠。
arXiv:2608.21867v1 Announce Type: new Abstract: LLM agents are moving from single-prompt use to long task streams in which reusable memory becomes a c…
12
Show HN: Poka-Yoke – mistake-proofing skills for Claude Code, with a benchmark
为Claude Code打造防错技能，经591次盲测验证，六类模型均显著提升。
Article URL: https://github.com/rainmanjam/poka-yoke Comments URL: https://news.ycombinator.com/item?id=49427559 Points: 1 # Comments: 0
13
TRACE: A Self-Evolving Skill Bank for Consistent, Limit-Aware LLM Agents
LLM代理落地关键在于一致性与边界感知，TRACE用自进化技能库破解可靠性难题，值得关注。
arXiv:2608.22793v1 Announce Type: cross Abstract: Reliable deployment of LLM agents in user-facing products depends not on raw task-solving ability bu…
14
The Compaction Cliff in Long-Running AI Agent Memory
安全规则与日志争抢上下文，压缩时规则一字之差即失效，长跑AI智能体记忆悬崖的实证研究
arXiv:2608.22752v1 Announce Type: new Abstract: A safety rule and an episodic log compete for the same tokens in an AI agent's context. When the budge…
15
From SQL Generation to Tool Selection: A Domain-Oriented Pattern for MCP Servers
LLM代理如何通过MCP高效触达企业数据？从SQL生成到工具选择的领域导向模式，是构建可靠数据接入的关键参考。
arXiv:2608.22063v1 Announce Type: new Abstract: Agents built on Large Language Models (LLMs) increasingly reach enterprise data through the Model Cont…
16
Show HN: Sloppie – Agentic development environment for Linux
4000行Python打造的Linux agentic开发环境，按分支保存任务，在目录中跑shell，轻量而强大。
Hi! In my own development work I have noticed I less and less reach for Emacs and rather need the combo of a coding agent + git diff viewer + a stack …
17
版本间隔 4 周 → 2 周：微软加速 WebView2 Runtime 更迭，和 Edge 浏览器同步更新
微软将WebView2 Runtime更新周期从4周压缩到2周，与Edge同步，开发者需留意更频繁的版本迭代。
IT之家 8 月 25 日消息，微软昨日（8 月 24 日）发布博文，宣布自 Microsoft Edge 152 版本（计划 8 月 27 日发布）开始，WebView2 Runtime 将跟随 Edge 浏览器的更新节奏， 从现有的 4 周一次缩短到每隔 2 周一次。 WebView2 Runt…
18
LLM Tool Failures: Only 3 Root Causes – Value, Condition, Intent
大模型工具调用总翻车？三大根因一次讲透，从“填表人”视角避开90%的坑。
Article URL: https://github.com/Jang-woo-AnnaSoft/execution-state-preflight/blob/main/who-fills-in-the-form.md Comments URL: https://news.ycombinator.…
19
Toward Effective and Reliable LLM Agents via Dynamic Ontology
用动态本体给智能体装上知识骨架，可靠性与效率双提升！
arXiv:2608.22974v1 Announce Type: new Abstract: Large language model (LLM) agents rely heavily on knowledge encoded in model parameters or presented a…
20
I Gave ChatGPT Access to My iMessage Conversations, and It Was More Useful Than I Expected
把iMessage交给ChatGPT，意外收获：聊天记录变私人助理，Mac用户即刻可试。
Getting ChatGPT connected to your Apple Messages comes with privacy trade-offs—but it has its uses.