AI daily brief

Agent-first apps need a home that proves the agent has already been working before the user asks anything, and a chat that makes the agent feel like it is doing real work, not just returning text. This skill ships both halves as one drop-in screen:

• Greeting — "Good evening, Alex." reveals glyph by glyph (rise + blur, TextRenderer) above a one-line summary of what the agent found.
• Story strip — a "Daily brief" circle (the flowing agent orb inside a breathing, turning accent ring while a new brief is unread) followed by company spotlight circles.
• Market pulse carousel — full-width glass cards (title, ±% pill, three big stats, a Swift Charts area + line sparkline) that snap one card at a time and auto-advance every 5 s, with capsule page dots.
• Quick entries — two tiles plus a tall highlight card (a wobbling gift icon, a "New" badge) for a daily game or promo.
• Podium — top three people on gold / silver / bronze steps with a Week / Month capsule picker (matchedGeometryEffect).
• Feeds — a news carousel ("Talent pulse") with scroll-transition scaling, four "Signals for you" rows with kind badges, and "Talent moves" cards (from-company → to-company).
• Floating agent bar — orb (with a thinking arc while a task runs), agent name, a status line that rotates every 3.5 s with a push transition, voice and inbox buttons with an unread badge. Tapping it zoom-transitions into the chat.
• Agent chat — "Hi, I'm Sage" header, date separators, Markdown bubbles (paragraphs, bullets, numbered lists, inline bold), the user's bubble in an accent gradient, a copy / thumbs up / thumbs down / regenerate bar under the latest agent text, task cards ("Needs you", spinning dashed active step), match list cards (photo tiles with % pills, "Save all"), staggered follow-up suggestions, one-tap glass prompt chips over a glass composer (send / stop / mic states).
• Thinking → streaming — your responder reports phases ("Reading your playbook" → "Scanning 2,418 profiles" → "Ranking by fit") into a shimmering thinking bubble; the reply then streams in word by word, the people card slides in 250 ms later, and new suggestions animate in.

The brand, backend and persistence are stripped: the feed, the agent and the conversation come in as plain models, every tap goes out through SWBriefActions, and replies come from one async closure.