正在打开每日关注
正在打开每日关注
用多技能交叉审计界面，汇总成可演示、可逐项审批并回传决策的 HTML 工件。
This little skill-chaining and approval/denial artifact flow is really doing it for me lately. Full details below 👇 (You can mix and match the particular skills obviously. Just loving the ones that Emil has been putting out most recently!) 1️⃣ Install @emilkowalski skills into your project (or globally). 2️⃣Copy/paste: "Run an audit on this [app/page/website] using the emil-design-eng skill. Then do another audit with the apple-design skill. Then spawn a separate agent acting as a senior design engineer that reviews the audits and collates them into an HTML artifact for human consumption. That HTML artifact should visually demonstrate the current state and the suggested improvement for each suggestion, and allows me to approve, deny, or discuss each one with you. At the end/bottom of the document, there should be a button that allows me to copy back all the decisions the I made after I've gone through the audit review artifact to implement changes or discuss certain ones further."
是什么 这条推文分享的不是某个具体 UI 组件，而是一套把设计审查变成“可执行决策”的 Agent 工作流：先分别调用 emil-design-eng 与 apple-design 对页面做审计，再让独立的资深设计工程师汇总为 HTML 工件。每条建议同时呈现当前实现、提案实现与可操作演示，人在阅读后逐项批准、拒绝或保留讨论，并在底部复制决策。
核心设计 关键是把意见从长文本变成证据闭环。截图中的 “Gold CTAs...” 通过代码差异和 Contact 按钮展示按压反馈；“Mobile nav...” 则用两个可切换的手机面板说明时长、曲线与体感。优先级、来源标签、Staff Note 和固定底栏将问题、理由、状态与下一步串起来。
关键洞察 多技能交叉审计减少单一审美视角，独立汇总代理承担冲突消解，人类审批则把 Agent 输出接入研发节奏。HTML 是低门槛、可分享、可回看且能承载交互的评审媒介，比静态报告更接近真实体验。
边界与风险 这是流程范式而非质量保证：技能的审美偏差、重复建议和错误代码判断仍需人工校验；演示若与生产组件脱节会制造虚假信心。还应保存审计版本、决策身份和实现状态，并在移动端、键盘操作及 reduced motion 下复验。