agent

前几天准备打鲲鹏杯，不巧的是正好那天在赶路没法好好打，就只能委托agent帮我解一下题写wp给我学习 比赛公告说了是禁ai的，所以我也没交flag，只是好奇为什么群里一直有ban队伍的提示，而且居然还有作弊队伍使用的模型名？ 赛方最后也公开的检测手段，我发现我的agent也中招了 复现一下学习学习 以下部分内容摘自鱼影安全公众号 https://mp.weixin.qq.com/s/i_Um5CC

一句正当的工期要求，就能让自主Agent无限循环。实测45个token的注入，把单次任务放大到3712万token、94.58美元，折合4.73个月订阅额度。附复现靶场与五层防御。

各位 Buffer 周末好，以下是本周「FreeBuf周报」，我们总结推荐了本周的热点资讯、一周好文，保证大家不错过本周的每一个重点！

OpenAI推出Daybreak，依托多AI Agent协同自动打通Web应用渗透测试、漏洞验证、补丁修复与复测全链路安全流程。

多款AI黑客框架运行在公网可访问服务器上，部分控制台未设身份认证，可被攻击者用于自动化渗透窃密，存在极高安全风险。

Enforcement defenses against prompt injection, such as CaMeL, Progent, IPIGuard and DRIFT, ask the agent's own model to write a security artifact (a program, a privilege policy, a dependency graph or

Mobile GUI agents read the screen of a phone and operate apps on a user's behalf to complete multi-step tasks such as online shopping or messaging. Because the agent acts on whatever the screen shows,

On-device inference requires kernels that make effective use of local accelerators. Efficient training and evaluation of models that generate Apple Metal kernels benefit from native macOS tools and co

Decision models answer a question with a typed output, such as a choice among options with a probability for each, so one model can choose an agent's actions, evaluate actions already taken, or score

Program evolution provides a promising approach to discover agent memory designs and study the choices underlying their effectiveness. We introduce MAPE, a framework that evolves whole memory programs

To keep long interaction contexts tractable, stepwise group-relative training for multi-turn LLM agents materializes each source trajectory into multiple response rows and valid tokens. This expansion

When language models revise a decision after another agent disagrees, neither reversal counts nor final accuracy shows whether they are fixing mistakes or abandoning correct answers. We introduce a co

Built primarily on OpenAI models, Amplitude’s proactive product agent, Wave, identifies what to build, helps teams ship changes with coding agents, and measures the impact.

Hey HN readers! I'm snipedbywifi ( https://gitlab.com/snipedbywifi ), and i've been working on terminactory. Please try it out if you'd like to play a cozy factory game while waiting for long agent ta

Secure, fast, efficient, battle-tested at Alibaba's scale. Hybrid architecture code review tool: deterministic pipelines + LLM Agent, precise line-level comments, built-in multi-language ruleset (NPE,

Keep your coding agent up to date with the latest web best practices https://goo.gle/modern-web-guidance Modern Web Guidance (Source) This repo is the source repository for GoogleChrome/modern-web-gui

Chrome DevTools for coding agents https://developer.chrome.com/docs/devtools/agents Chrome DevTools for agents Chrome DevTools for agents ( chrome-devtools-mcp ) lets your coding agent (such as Antigr

现在让 Agent 写业务代码、跑单测、修线上 Bug ，基本已经没什么新鲜感了。有编译器、类型检查和报错堆栈兜底，它自己多试几轮总能收敛。 最近几个月换了个更折磨它的场景：把 MinePilot 当实验田，试试看能不能让 Agent 在原生 Minecraft 里稳定造出十几万到三十几万方块、玩家真能走进去逛的大型建筑。 我原本以为跟以前做策略 DSL 一样，先给它造一套带编译校验的底座就够了（

我让 Claude Opus 5.5 跑整个系统的全项目测试，测试到有一个比较极限的场景有 BUG ，于是我让 Agent 自己提了一个 jira 工单，然后呼叫对应模块的 owner 来解决这个 bug ，但是 owner 告诉我，他看不懂我的工单里是在说什么 于是我查看了一下 Agent 生成的工单，工单写的十分详细，但是根本没有重点，只是把我们做过的实验全部放上去，并且自创了一堆词汇，这些词

Runner 是我做的一个管理 CLI agent 的开源桌面工具，之前在 V 站 介绍过 ，可以在同一个 app 里用 Claude Code 、Codex 等，也可以把不同的 agent 放进一个 crew 里一起做任务。上次发 0.18 桌面端远程控制 的时候提过移动端会优先做 iOS ，最近把 Runner iPhone 版的 UI 和主要交互定下来了，已经上传 TestFlight ，开

Claude 开了 5x ，124 刀，用了半个月，昨天被封了，申请退款，退了 56 刀。 GPT 已经彻底废了，完全干不了活，x 上看到不少推荐 Gemini 的，模型能力还可以，5x 土区大概 280 元，之前有 20x 大概 470 元。 但是众所周知，谷歌反重力并不好用。似乎没有办法将它接入到 Claude code 或者 codex 。