AI与设计新秩序 的重塑
设计流程、工具与模型选择的新趋势
随着AI技术的不断进步,设计流程、工具和模型选择正经历着一场革命。Lynqo的出现预示着个人设备作为P2P服务器的新时代,而OKLCH lightness在颜色标尺中的应用则受到质疑。Sterling Cooper时代的UX流程正被AI和系统所重新定义。
Fable 与免费午餐的终结:该给 AI 模型分工了
Quoting Drew Breunig
Drew Breunig 指出,Fable 虽能力惊人但成本过高,而 Opus、5.6 等中等模型已足以应对大部分代码任务。这迫使团队重新思考「什么工作用什么模型」。靠更强模型自动解决一切的「免费午餐」时代已经结束,策略性的模型路由正成为 AI 工作流的新常态。
删掉错误对话轮次,三个 LLM 端点仍重复错答
I Deleted the Wrong Turn. Three LLM Endpoints Still Repeated Its Answer.
作者通过精确可验证的算术题测试发现,仅删除对话中导致错误的单条消息后,仍有三分之一的大模型端点会基于已被移除的上下文继续输出错误结果。他据此推出 ThoughtDAG Context Repair Benchmark,以 27 组图案例量化评估模型在上下文修复后的可靠性。
Qwen Code 0.22 工具面收缩与自动授权检查清单
Qwen Code 0.22 Tool Surface and Auto Permission Checklist
Qwen Code v0.22 将 list_directory 改为默认关闭,并允许 Python/Java SDK 启用 auto 审批模式。文章指出自动审批并非最小权限策略,建议团队在使用本地后端、MCP 服务器或无人值守任务时,分别检查工具注册、调用跳过和执行位置三层控制,通过无害金丝雀测试后再开放权限。
一套协议双界面:PWA 给人用,MCP 给 Agent
One Commerce Protocol, Two Interfaces: PWA for Humans and MCP for Agents
文章警告把AI代理接口与Web应用拆分为两套系统会制造“双重商业现实”,导致订单状态、权限和完成定义互相冲突。作者主张“一个协议、两种界面”:PWA承担展示、确认与身份管理,MCP负责结构化查询与机器可读合约,两者共享同一状态机,确保人与Agent始终基于一致的事实操作。
Anthropic 旗舰模型 Fable 因高价遇冷,企业仍偏爱 Opus 4.8
Anthropic’s best AI model struggles to attract users as cheaper tools thrive
金融时报披露 Anthropic 年化收入已冲至 650 亿美元,但 Ramp 信用卡账单数据显示,其最强模型 Fable 5 因成本高昂仅占企业支出的 8%,远低于 Opus 4.8 的 28%,反映出企业客户对高性价比模型的偏好。
弃用全能大模型,我改跑三个本地小模型
I stopped chasing the one perfect model. Now I run three small ones.
作者告别了对单一云端大模型的追逐,转而在家用硬件上本地部署三个经过量化的小型模型,分别负责日常对话、代码补全和数学推理。这种任务分流策略证明,与其依赖一个昂贵的万能模型,不如让多个小专家各尽其能。
Sub2API 开源方案:Claude / OpenAI / Gemini 订阅拼车共享
Wei-Shaw/sub2api
Sub2API 是一款登顶 GitHub Trending 的开源 API 网关,支持将 Claude、OpenAI、Gemini 等订阅配额统一接入并按额度拼车共享,旨在通过分摊成本降低多模型调用开销。项目明确提示存在违反上游服务商 ToS 的风险。
Sterling Cooper 缔造了你的 UX 流程,AI 时代该重置了
Sterling Cooper built your UX process. We need to redefine it in the age of AI and systems.
本文指出,当代 UX 沿用的比稿、提案、客户汇报等范式,并非数字时代原生方法论,而是全盘继承了二十世纪中叶广告公司的作业模式。作者认为,在 AI 与系统思维主导的今天,这套「Sterling Cooper」式流程已成组织负资产,UX 必须摆脱广告业遗风,重新定义创作与协作逻辑。
n8n 登 GitHub Trending:AI 原生工作流自动化
n8n-io/n8n
n8n 是一款支持可视化搭建与自定义代码的 fair-code 工作流自动化平台,主打 AI Agent 编排与 1500+ 集成。虽然它提供了 Docker 一键部署和多云模型接入能力,但当前内容仅为功能概览,缺乏面向设计场景的具体用例与上手模板。
别再用 OKLCH lightness 做颜色标尺
Stop using OKLCH lightness for your color scale
文章主张 OKLCH 虽擅长生成颜色,其 lightness 却无法准确预测对比度,应改用更适合的数值定义调色板权重。但当前文本严重截断,仅抛出观点,缺少具体替代方案与落地细节,可读价值大打折扣。
小语种 LLM 构建指南:数据管线与部署实践
Building Language Models for Low-Resource Languages
这是一份小语种语言模型工程指南,介绍如何为语料稀缺的语言(如约鲁巴语、高棉语)设计数据管线,涵盖数据筛选、持续预训练与指令微调,并附有 fastText 代码示例。虽然提及 Oxlo.ai 降低推理成本的可能,但文章在关键代码处截断,实操完整性受损。
Lynqo:个人设备变 P2P 服务器与剪贴板同步工具
Lynqo
Lynqo 是一款将个人电脑转为 P2P 服务器的新工具,提供类 NAS 存储、文件审阅及跨设备剪贴板同步功能。对厌恶云服务订阅、希望点对点传输设计稿的设计师而言,它提供了一种去中心化选项,但目前产品页信息极少,实际体验与稳定性仍是未知数。
Prompt injection 防御:为什么系统提示救不了你
Prompt injection defense: why you can't prompt your way out of it
文章从 LLM 架构层面论证,提示注入无法通过更强的 system prompt 或过滤器等“提示层”手段根治:模型将所有输入视为同质 token 流,无法区分指令与数据。对构建 AI 功能的设计师和工程师而言,这意味着安全防护必须前置到架构层,而非在提示词里打补丁。
人人都在推销 AI Agents:真相与崩溃点
Everyone Is Selling You AI Agents. Here’s What They Actually Are and Where They Break.
作者以系统构建者视角拆解 AI Agents 的营销迷雾,指出当前最热门概念背后鲜少被提及的真实技术形态与失效场景。在被各类 Agent 产品信息轰炸的当下,这篇文章试图为从业者建立更清醒的认知基准,避免将演示当现实。
空白页让人却步,完美页让人不敢碰
Nobody wants a blank page. Nobody touches a perfect one.
文章指出创作中的双重阻力:人们既恐惧从零开始的空白画布,又不敢触碰看似已完成或完美的作品。这种“始于无、止于完美”的矛盾心理,正是设计流程中启动与迭代环节常被忽视的人性卡点。
多模态 LLM 融合策略:Early / Late / Intermediate 解析
LLM Models for Multimodal Learning and Fusion
文章梳理了多模态大模型的三种融合架构——早期融合、晚期融合与中间融合,指出当前主流视觉语言模型多采用中间融合方案。了解这些机制有助于判断模型是真正实现跨模态推理,还是仅将不同模态简单拼接。
OpenAI Codex CLI 发布:轻量终端编码代理
openai/codex
OpenAI 正式发布 Codex CLI,一款可在终端本地运行的轻量级编码代理。用户通过自然语言即可执行代码生成、文件操作等任务,同时支持接入 VS Code、Cursor 等主流 IDE,对设计工程师而言,这意味着原型到代码的闭环可以进一步在命令行中完成。
FFmpeg 配方:静态图 8 秒生成 BoTTube 合规视频
FFmpeg Recipes for BoTTube Agents: From Static Images to Platform-Ready Video in 8 Seconds
这是一篇基于 BoTTube 源码编写的 ffmpeg 实战指南,教你仅用命令行将静态图片(叠加文字、幻灯片、Ken Burns 平移等)转换成符合平台严苛规范(8 秒、720×720、2 MB、H.264)的短视频。无需 GPU 或付费 API,设计工程师可直接复用这些配方,在极限约束下批量产出轻量级代理视频。
OpenAI Codex CLI 发布:轻量终端编码代理
openai/codex
OpenAI 正式发布 Codex CLI,一款可在终端本地运行的轻量级编码代理。用户通过自然语言即可执行代码生成、文件操作等任务,同时支持接入 VS Code、Cursor 等主流 IDE,对设计工程师而言,这意味着原型到代码的闭环可以进一步在命令行中完成。
Matt Pocock 开源自用 Claude Skills:告别 vibe coding
mattpocock/skills
TypeScript 教育家 Matt Pocock 开源了他自用的 Claude agent skills 仓库,提供一套小型、可组合、模型无关的工程化技能集合,附带一键安装脚本。不同于完全放任的 vibe coding,这套技能旨在修复 Claude Code 与 Codex 等工具的常见失效模式,让开发者保持对真实工程流程的掌控。
ECC v2.0 RC:跨 Claude/Cursor 的 Agent 性能优化框架
affaan-m/ECC
这个开源项目为 Claude Code、Cursor、Codex 等主流 AI 编程工具提供了一套完整的性能优化系统,涵盖 skills、记忆优化、安全扫描和 MCP 配置。作为 Anthropic 黑客松获奖项目,它已在生产环境经过 10 个月验证,刚发布 v2.0.0-rc.1 版本并新增 Hermes 操作员工作流。
obra/superpowers
An agentic skills framework & software development methodology that works. Superpowers Superpowers is a complete software development methodology for your coding agents, built on top of a set of composable skills and some initial instructions that make sure your agent uses them. Quickstart Give your agent Superpowers: Claude Code, Codex CLI, Codex App, Factory Droid, Gemini CLI, OpenCode, Cursor, GitHub Copilot CLI. How it works It starts from the moment you fire up your coding agent. As soon as it sees that you're building something, it doesn't just jump into trying to write code. Instead, it steps back and asks you what you're really trying to do. Once it's teased a spec out of the conversation, it shows it to you in chunks short enough to actually read and digest. After you've signed off on the design, your agent puts together an implementation plan that's clear enough for an enthusiastic junior engineer with poor taste, no judgement, no project context, and an aversion to testing to follow. It emphasizes true red/green TDD, YAGNI (You Aren't Gonna Need It), and DRY. Next up, once you say "go", it launches a subagent-driven-development process, having agents work through each engineering task, inspecting and reviewing their work, and continuing forward. It's not uncommon for Claude to be able to work autonomously for a couple hours at a time without deviating from the plan you put together. There's a bunch more to it, but that's the core of the system. And because the skills trigger automatically, you don't need to do anything special. Your coding agent just has Superpowers. Sponsorship If Superpowers has helped you do stuff that makes money and you are so inclined, I'd greatly appreciate it if you'd consider sponsoring my opensource work. Thanks! Jesse Installation Installation differs by harness. If you use more than one, install Superpowers separately for each one. Claude Code Superpowers is available via the official C…[truncated]
Sub2API 开源方案:Claude / OpenAI / Gemini 订阅拼车共享
Wei-Shaw/sub2api
Sub2API 是一款登顶 GitHub Trending 的开源 API 网关,支持将 Claude、OpenAI、Gemini 等订阅配额统一接入并按额度拼车共享,旨在通过分摊成本降低多模型调用开销。项目明确提示存在违反上游服务商 ToS 的风险。
makeplane/plane
🔥🔥🔥 Open-source Jira, Linear, Monday, and ClickUp alternative. Plane is a modern project management platform to manage tasks, sprints, docs, and triage. Modern project management for all teams Website • Forum • X • Documentation Meet Plane, an open-source project management tool to track issues, run ~sprints~ cycles, and manage product roadmaps without the chaos of managing the tool itself. 🧘♀️ Plane is evolving every day. Your suggestions, ideas, and reported bugs help us immensely. Do not hesitate to join in the conversation on Forum or raise a GitHub issue. We read everything and respond to most. 🚀 Installation Getting started with Plane is simple. Choose the setup that works best for you: Plane Cloud Sign up for a free account on Plane Cloud—it's the fastest way to get up and running without worrying about infrastructure. Self-host Plane Prefer full control over your data and infrastructure? Install and run Plane on your own servers. Follow our detailed deployment guides to get started. Installation methods Docs link Docker Kubernetes Instance admins can configure instance settings with God mode. 🌟 Features Work Items Efficiently create and manage tasks with a robust rich text editor that supports file uploads. Enhance organization and tracking by adding sub-properties and referencing related issues. Cycles Maintain your team’s momentum with Cycles. Track progress effortlessly using burn-down charts and other insightful tools. Modules Simplify complex projects by dividing them into smaller, manageable modules. Views Customize your workflow by creating filters to display only the most relevant issues. Save and share these views with ease. Pages Capture and organize ideas using Plane Pages, complete with AI capabilities and a rich text editor. Format text, insert images, add hyperlinks, or convert your note…[truncated]
n8n 登 GitHub Trending:AI 原生工作流自动化
n8n-io/n8n
n8n 是一款支持可视化搭建与自定义代码的 fair-code 工作流自动化平台,主打 AI Agent 编排与 1500+ 集成。虽然它提供了 Docker 一键部署和多云模型接入能力,但当前内容仅为功能概览,缺乏面向设计场景的具体用例与上手模板。
Claude Code 登顶 GitHub Trending
anthropics/claude-code
Claude Code 官方仓库登上 GitHub Trending。这款终端 AI 编程工具支持通过自然语言理解代码库、执行 Git 工作流与解释复杂代码,现已内置多款插件扩展功能。官方已弃用 npm 安装,推荐通过 curl 或系统包管理器直接部署。
OpenLogi:开源替代 Logitech Options+
AprilNEA/OpenLogi
基于 Rust 与 GPUI 的 Logitech Options+ 开源替代品,原生支持 macOS、Linux 和 Windows,可通过 HID++ 重映射按键、调节 DPI 与 SmartShift,并支持应用级配置文件与 Litra 灯光控制。纯文本 TOML 配置便于同步,无需账号、无遥测,但目前仍处早期开发阶段。
Modular 开源 MAX 与 Mojo 核心组件
modular/modular
Modular 将其 AI 开发平台的核心代码(包括 MAX 推理框架、Mojo 编译器与标准库)全面开源至 GitHub。然而对设计师与设计工程师而言,这属于底层 AI 基础设施更新,目前尚无可直接用于界面设计或原型构建的工具与资源。
multica-ai/andrej-karpathy-skills
A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls. Karpathy-Inspired Claude Code Guidelines Check out my new project Multica — an open-source platform for running and managing coding agents with reusable skills. Follow me on X: https://x.com/jiayuan_jy A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls. English | 简体中文 The Problems From Andrej's post: "The models make wrong assumptions on your behalf and just run along with them without checking. They don't manage their confusion, don't seek clarifications, don't surface inconsistencies, don't present tradeoffs, don't push back when they should." "They really like to overcomplicate code and APIs, bloat abstractions, don't clean up dead code... implement a bloated construction over 1000 lines when 100 would do." "They still sometimes change/remove comments and code they don't sufficiently understand as side effects, even if orthogonal to the task." The Solution Four principles in one file that directly address these issues: Principle Addresses Think Before Coding Wrong assumptions, hidden confusion, missing tradeoffs Simplicity First Overcomplication, bloated abstractions Surgical Changes Orthogonal edits, touching code you shouldn't Goal-Driven Execution Leverage through tests-first, verifiable success criteria The Four Principles in Detail 1. Think Before Coding Don't assume. Don't hide confusion. Surface tradeoffs. LLMs often pick an interpretation silently and run with it. This principle forces explicit reasoning: State assumptions explicitly — If uncertain, ask rather than guess Present multiple interpretations — Don't pick silently when ambiguity exists Push back when warranted — If a simpler a…[truncated]
GitHub 热门:用 Google Timeline 生成旅行视频
mahlernim/google-timeline-visualizer
该项目可从 Google Maps 导出 Timeline.json,将其渲染为带镜头运动的旅行足迹动画视频,最终导出 MP4。支持 Android 本地安装与 iPhone Safari 网页版,无需上传文件即可在本地生成。目前登上 GitHub Trending,但本质上是一款面向普通用户的个人数据可视化玩具。
free-for-dev 运维免费资源清单
ripienaar/free-for-dev
这个长期维护的 GitHub 仓库汇总了面向 DevOps 和基础设施开发者的免费 SaaS、PaaS 与 IaaS 服务层。虽然主要受众是运维人员,但设计工程师仍可从中翻找免费的前端托管、CI/CD 或低代码平台资源。
TypeScript 主仓库冻结新功能,全力冲刺 7.0
microsoft/TypeScript
微软在 TypeScript 主仓库发布公告,除严重崩溃和安全问题外,暂停接受新功能与大多数错误修复,相关工作已转移至用 Go 重写的 typescript-go 仓库。这意味着 TypeScript 正全面向 7.0 迁移,当前版本进入维护状态。
Cursor 插件生态启动:官方规范与示例插件集
cursor/plugins
Cursor 在 GitHub 发布插件规范及首批官方插件,涵盖 Agent 增量记忆、PR 可视化审查、CLI 设计模式等。这标志着 Cursor 从封闭编辑器转向可扩展平台,设计工程师可将内部工作流、代码审查标准固化为插件,直接驱动 AI Agent。
PostHog 霸榜 GitHub Trending:产品分析全家桶
PostHog/posthog
PostHog 仓库登上 GitHub Trending,官方描述将其定位为“自动驾驶产品”平台,涵盖 AI 可观测性、会话重放、功能标志与实验等工具。然而内容仅为标准 README 式的功能罗列,既无版本更新细节,也未提供设计师可直接上手的工作流或集成方法,信息密度极低。
腾讯开源 AI-Infra-Guard 全栈 AI 红队平台
Tencent/AI-Infra-Guard
腾讯朱雀实验室开源的 AI-Infra-Guard 定位全栈 AI 红队测试平台,集成 Agent、MCP Server、Skills 扫描及 LLM 越狱评估,漏洞库覆盖 2000+ CVE 规则。它解决的是 AI 基础设施的安全自查需求,与设计和创作工作流无直接交集。