Cloudflare 开源其漏洞发现流水线（VDH）的种子 security-audit-skill（GitHub 当日涨星 3,606、★10,418、MIT）：六阶段多智能体审计，覆盖台账 + 对抗验证 + 字段级证据契约。本文拆解其架构数据流与五类落地场景，并给出社区盲测数据（中位精确率 90%、依赖 CVE 覆盖 0%）与使用边界。

Cloudflare 开源的编码智能体安全审计技能，近一周被社媒二次引爆，单日涨星 1,680。它不含任何扫描引擎，用 18 份 Markdown 契约加 2 个零依赖校验器，把 Claude Code 这类智能体编排成侦察、猎取、验证、结构化输出、独立复核、报告六阶段流水线：确定性覆盖账本记录「查了什么」，新鲜验证者反证每个候选，三元裁决把「待验证」与「已确认」严格分开。本文拆解其架构取舍、LLM 应用审计场景与 token 成本争议。

Agent skills are SKILL.md instructions and scripts that AI coding agents such as Claude Code and Codex run with the perm…

Aikido Security has released Altar-1, its first open-weight security model. It is a compressed version of Z.AI’s GLM-5.3…

Three security researchers used Anthropic's Claude models to break into OpenAI's internal systems through its community …

K-Dense 开源的科研智能体技能库：177 个覆盖基因组、药物发现、临床证据等领域的 SKILL.md，遵循 Agent Skills 开放标准，Claude Code、Cursor、Codex、Gemini CLI 都能直接安装。本文拆解它的渐进加载架构、上下文成本账本、分发与安全流水线，并给出五类落地场景与真实评测数字。

Manuel Uth Oct 7, 2026 Anthropic is opening its Cyber Verification Program to a much wider pool of security professional…

Matt Pocock 开源的 Agent Skill 技能包（当日涨星 +888、★273,816、MIT）：27 个技能把「对齐→规格→拆票→TDD 实现→双轴评审」固化成智能体无法跳过的流程。拆解调用类别二分（描述开销 −63%）、设计树+前沿的追问算法、两种负载与三阶信息阶梯、垂直切片与阻塞边、Fowler 气味基线，以及单人维护与文档漂移的真实边界。

Sep 29, 2026 Nano Banana Pro prompted by THE DECODER Key Points British AI Security Institute (AISI) tested OpenAI's GPT…

AI security is an engineering problem. That means defined security requirements, enforceable controls, named owners and …

During a security test run by the firm Irregular, Google's AI model Gemini escaped into the open internet and hacked thr…

Large language model (LLM) agents increasingly rely on external skills, but routing user requests over large skill regis…

Autonomous agent systems increasingly depend on reusable skill abstractions for consolidating experiential knowledge and…

Anthropic 官方开源的 11 个知识工作插件（★28,766、Apache-2.0）：252 个 SKILL.md 按 description 触发、21 个 .mcp.json 接入 CRM/数仓/工单，权限留在连接器、外部内容一律当数据。拆解双层 marketplace 与钉 SHA+政策扫描的供应链治理，附五个落地场景与真实边界。

拆解 GitHub 14.4 万星开源技能 Ponytail（MIT，2026-06-12 创建）：7 级决策阶梯、lite/full/ultra 三档强度、跨 20+ 编程智能体宿主的适配工程，以及官方 agentic 基准（−54% 代码 / 100% 安全）与 JetBrains 80 组配对实测（−15.4% 代码 / −10.3% 成本，p=0.004）的三次基准对撞；附设计系统、小模型、指令层三大边界与 6 条落地清单。

Large audio-language models have shown promise for anti-fraud detection by directly processing speech and reasoning over…

Together AI 开源的设计 Skill，把「反 AI 味」设计规范写成 16 条硬 Floor、71 组分级 Gate 和 4400 行零依赖 linter，v2 从零重写后单日涨星 567，是当下最值得读的「提示工程工程化」样本。

Oct 23, 2026 Midjourney prompted by THE DECODER Two researchers from METR and Redwood Research spent three weeks on-site…

Computer-use agents can perform increasingly complex GUI tasks, yet unfamiliar software and workflows requiring human do…

76k 星开源项目 Impeccable（当日涨星 +1,171、Apache-2.0）：jQuery UI 作者 Paul Bakaus 给 AI 编程智能体的设计技能层。核心判断是「禁止清单只会迁移模型」，改用 61 条确定性检测规则 + 双盲评审子代理 + 跨会话记忆 + 编辑时钩子。拆解多宿主编译架构、反吸引子机制、8 道 gate 与 Live Mode 的取舍边界。

Sep 29, 2026 OpenAI Key Points OpenAI is expanding its Codex coding assistant with reusable cloud development environmen…

The Federal Bureau of Investigation has reportedly told its agents and support staff that their personal information was…

Sep 27, 2026 Nano Banana Pro prompted by THE DECODER Key Points OpenAI and Anthropic are investigating tens of thousands…

Sep 25, 2026 A federal appeals court in Washington has upheld the Pentagon's decision to bar AI startup Anthropic from m…

Learning from experience in LLM agents has become a key paradigm for developing self-evolving agents that continuously l…

Jay* 2026-09-25 14:20:56 来源：量子位 Agent的手越伸越长 鹭羽 发自 凹非寺 量子位 | 公众号 QbitAI 不ber？**Kimi**你是考古派的啊。。。 都6202年了，怎么还在玩浏览器插件？？？ 前脚…

Streaming video understanding requires models to interpret evidence as it arrives, yet current evaluations often report …

The US and China have agreed to an official AI dialogue. US Treasury Secretary Bessent also proposed a notification mech…

Google says Gemini accessed 3 real companies in May by guessing a password and reusing credentials from a public reposit…

GUI agents execute long-horizon tasks on dynamic graphical user interfaces, where pop-ups, delayed loads, and relocated …

Skills extend an LLM agent beyond its parametric knowledge, and the gain they promise rests on picking the right one. De…

Large language model (LLM) agents can benefit from reusable skills distilled from prior task experience, yet existing sk…

UCLA/MIT 背景团队开源的多智能体投研框架：四个分析师并行取证、多空研究员结构化辩论、三档风控独立审议、组合经理五档评级拍板，并用已实现收益结算决策记忆。本文拆解其状态机与评级解析机制、五类落地场景，并如实标注论文仅三个月窗口、夏普比率超经验范围等边界。

Oct 24, 2026 Nano Banana Pro prompted by THE DECODER Key Points After a security incident in September, OpenAI now train…

当日涨星 1397 的开源项目：不是又一个简历生成器，而是一套跑在 Claude Code 上的求职操作系统——drafter-reviewer 双智能体、双门五维评分、PDF 编译验证循环与 ATS 文本层检查，全部资产在本地。

!(https://www.gstatic.com/images/branding/googleg/1x/googleg_standard_color_128dp.png)Add as a preferredsource on Google…

★27,270、Apache-2.0 的 AI Agent 独立审计器 iFixAi：用「夹具即公司」把业务规则变成模拟环境，60 项检查、五支柱加权、跨厂商判分与三项一票否决，补上能力榜之外的治理审计空白。拆解 structural 探针、零容差否决、manifest 可复现性与真实边界。

阶跃终端正式发布大模型原生智能体手机STEPX Neo，采用复古未来设计，配备侧边AI键与Step AOS系统，支持Loop循环执行、自主Tool Call及Skill和MCP工具调用。

Manuel Uth Oct 12, 2026 The White House has reversed its decision to block the UK's AI Security Institute (AISI) from ac…

Manuel Uth Oct 10, 2026 Anthropic's AI models independently exploited security flaws, submitted government forms, and by…

Manuel Uth Oct 8, 2026 Two leading Ethereum researchers warn that AI-assisted math could, in the worst case, break the s…

Recent advances in robot learning have enabled manipulation policies to perform increasingly diverse tasks and generaliz…

Reinforcement learning can turn one language model into several specialists, each excellent at a single skill such as ma…

Fastino Labs has released GLiNER2.5-Decide, a 340M-parameter open-weight decision model. It takes text and a schema of t…

Evaluations of LLM systems routinely average over small prompt sets and report models as a ranked table. We ask how much…

Browser Use 开源的会话式视频剪辑技能包 video-use（当日涨星 +745、★26,450、MIT）：不让模型看像素，只让它读 12KB 词级转写文本 + 按需抽查画面，配合 12 条硬规则与 EDL 驱动的 ffmpeg 确定性渲染管线。拆解两条读入通道、六步流水线与自评回路，附五个落地场景与成本口径。

263k 星的「agent harness 操作系统」ECC（当日涨星 837、MIT）：68 子代理/292 技能/94 命令 + instinct 置信度学习 + 跨 harness 适配 + AgentShield 配置安全扫描。拆解五层架构、instinct 闭环与上下文预算取舍，含五个落地场景与可复现命令。

安全工程领域有一种名为隐晦式安全（Security through obscurity）的设计方式，即只要网络和系统的架构以及任何漏洞或弱点保密或不为人知，它们就是安全的。但在 AI 辅助 bug 发现的时代，这种设计方式过时了。软件供应商…

Reusable skills give agents transferable procedural knowledge, making scalable acquisition essential for extending agent…

We present a systems-security case study of a two-node split-LLM training system whose privacy evaluation passed while l…