Building useful tools
for intelligent work.为智能工作
打造有用的工具。

A native voice-writing app for Apple silicon Macs that keeps its core Dictate and Translate paths on the device.面向 Apple 芯片 Mac 的原生语音写作应用，核心听写与翻译链路都在设备上完成。

• Quick Dictate delivers after a short silence; press-and-hold and follow-up voice edits remain available.Quick Dictate 会在短暂停顿后自动写入，同时支持按住说话与连续语音修改。
• Command can use an optional local MLX model or an OpenAI-compatible API configured by the user.Command 可选用本地 MLX 模型，或由用户配置的 OpenAI 兼容 API。
• No account, subscription, telemetry, or server-side transcript storage is required.无需账号与订阅，不含遥测，也不会在服务器存储转写内容。

A browser-native Chat, Local Agent Workspace, and Local Model Studio built on WebLLM, with no application backend or account.基于 WebLLM 的浏览器原生对话、本地 Agent 工作区与本地模型工作室，无应用后端与账号系统。

• The curated catalog covers 20 models, including three Hermes models with native function calling.精选目录包含 20 个模型，其中三个 Hermes 模型支持原生函数调用。
• Sandboxed local tools can inspect selected files, search context, calculate, and create artifacts after diff approval.沙箱化本地工具可检查选定文件、搜索上下文、计算，并在差异确认后创建产物。
• The advanced catalog groups 163 official WebLLM records into 65 logical models and selects compatible quantization.高级目录将 163 个官方 WebLLM 记录整理为 65 个逻辑模型，并自动选择兼容量化版本。

An open component system for moving from a complete product page to a focused interaction or precise motion behavior.一个开源组件系统，覆盖完整产品页面、聚焦式交互与精确动效行为。

• The registry publishes 4 complete page blocks, 48 product components, and 44 motion primitives and guides.Registry 发布 4 个完整页面区块、48 个产品组件，以及 44 个动效原语与指南。
• Motion, GSAP, Three.js, WebGL, SVG, and CSS implementations share one install path.Motion、GSAP、Three.js、WebGL、SVG 与 CSS 实现共用一套安装路径。
• The Agent Skill supports page building, recommendation, composition, implementation, review, and contribution.配套 Agent Skill 支持页面构建、推荐、组合、实现、评审与贡献。

A CLI, Agent Skill, and GitHub Action that runs Chromium and Lighthouse where the check is invoked and keeps reports local.一套 CLI、Agent Skill 与 GitHub Action，在执行检查的位置运行 Chromium 和 Lighthouse，并将报告保留在本地。

• Desktop and mobile evidence covers HTTP, page, console, asset, layout, metadata, and structured-data failures.桌面端与移动端证据覆盖 HTTP、页面、控制台、资源、布局、元数据与结构化数据问题。
• One report model renders HTML, Markdown, and JSON for people, agents, and CI.同一份报告模型可输出 HTML、Markdown 与 JSON，供人、Agent 与 CI 使用。
• Stable issue codes and severity thresholds let the CI command act as a release gate.稳定的问题代码与严重级别阈值让 CI 命令可以直接充当发布门禁。

A self-contained Agent Skill for giving products one ownable visual system with a disciplined creative and runtime process.一个自包含的 Agent Skill，通过完整的创意与运行时流程，为产品构建具有辨识度的视觉系统。

• Creative search, adversarial selection, and distance checks keep the chosen direction specific to the product.创意搜索、对抗式筛选与距离检查让最终方向真正属于当前产品。
• The method chooses among Canvas 2D, Three.js, raw WebGL, and SVG after the experience is defined.在体验方向确定后，再从 Canvas 2D、Three.js、原生 WebGL 与 SVG 中选择实现。
• Deterministic runtime QA captures responsive, reduced-motion, interaction, and lifecycle states for critique.确定性运行时 QA 会捕捉响应式、减少动态效果、交互与生命周期状态用于评审。

An image-making system that combines a searchable public reference library with reusable scene inputs and asynchronous generation.一套图像创作系统，将可搜索的公开参考库、可复用场景输入与异步生成连接在一起。

• The public gallery contains 1,000 curated prompt-image pairs for visual search and adaptation.公开图库包含 1,000 组精选提示词与图像，可用于视觉搜索与改写。
• The open gpt-image-2-search Skill lets agents find references without an API key.开源 gpt-image-2-search Skill 让 Agent 无需 API key 即可搜索参考。
• The product carries selected scenes through account, credit, queue, storage, and result workflows.产品将选定场景连接到账户、点数、队列、存储与结果工作流。