Three updates this week. The first one changes how Rebyte skills are distributed: they're now addressable by URL and embeddable anywhere. The other two — the Office chip and Show Me How — both close the gap between what an agent says and what you can actually use.

Run Any Skill With a Single Click — From Anywhere

Every Rebyte skill is now reachable from a single URL. The pattern is dead simple:

https://app.rebyte.ai/new?prompt=<url-encoded prompt>

Click it, sign in, and an agent boots in a cloud VM with the skill already loaded and the task already running. No CLI install, no API key, no setup. The same URL works from a README, a tweet, a Slack message, or a link in an email.

There are two embed formats. The first is a markdown badge for READMEs and blog posts:

[![Run on Rebyte](https://raw.githubusercontent.com/ReByteAI/run-any-skill-with-single-click/main/badge-v3.svg)](https://app.rebyte.ai/new?prompt=Use%20the%20YOUR_SKILL%20skill.%20YOUR%20PROMPT%20HERE)

The second is a one-line script tag for websites, landing pages, and docs:

<script src="https://rebyte.ai/embed.js"
  data-prompt="Use the YOUR_SKILL skill. YOUR PROMPT HERE."
  data-text="Run on Rebyte"
  data-theme="light">
</script>

We've seeded a gallery with community-built skills to show what's possible: github.com/ReByteAI/run-any-skill-with-single-click. It includes gstack from Garry Tan (a virtual engineering team running a structured sprint), last30days from Mark Van Horn (what the internet is saying about any topic), autoresearch from Udit Goenka (iterative code improvement loop), Slidev, claude-d3js, visual-explainer, webgpu-claude-skill, and more. Each one is a single click.

This is also why the email version of this post is different from any previous email we've sent. Every button in that email is the same embed — literally the feature in action.

The Office Chip: Word, Excel, PowerPoint, PDF

Rebyte now ships with a first-class Office chip. Type /office in any task and you get a gallery of four formats — Word, Excel, PowerPoint, PDF — each grouped into three categories: Create, Edit, and Refine.

Behind the chip are four dedicated skills — xlsx, docx, pptx, and pdf — each with hundreds of lines of task-specific knowledge. The Excel skill knows how to build a three-statement financial model with linked schedules and sensitivity tables. The Word skill knows SOW templates and track-changes editing. The PDF skill handles form filling, OCR on scanned docs, and page-level operations.

Every Office file the agent produces lands in your workspace Artifacts tab and previews inline via Google Drive — so you can see the sheet, deck, or document the moment it's written, without downloading anything. Share the artifact URL and your team sees the same preview.

The paradigm matters. You're not asking an agent to "maybe write a spreadsheet if it can." You're picking Office, picking a format, picking an action, and getting a polished file back. Every time.

Show Me How: Interactive Widgets in Chat

The second update is a new skill called Show Me How. It lets any agent generate interactive HTML widgets that render directly inside the chat — sandboxed iframes that you can click, drag, and explore.

Ask "show me how compound interest works" and you get a live calculator with sliders. Ask "chart my Q1 revenue by region" and you get a Chart.js bar chart. Ask "visualize the agent's memory flow" and you get an animated GSAP explainer with play/pause controls.

Every widget is vanilla HTML + CSS + JS — no React, no bundlers, no build step. The agent writes the source and the chat renders it. You can pick an aesthetic (editorial, blueprint, paper-ink, mono-terminal, and more) and the font system auto-resolves colors and typography via CSS variables, so widgets adapt to light and dark mode automatically.

Behind the scenes, Show Me How is opinionated about not looking AI-generated. No emoji headers. No three-column card grids. No gradient text. No glowing shadows. The skill explicitly teaches the agent what distinctive design looks like, with a squint-test quality gate before it delivers.

The point is that text alone is a lossy format for many questions. When someone asks how an algorithm works or what a dataset shows, a paragraph can describe it — but a widget lets you explore it. Now your agents can do both in the same reply.

Try All Three

All three updates are live now. Click any badge in the gallery to run a skill with a single click. Type /office for Microsoft Office files. Or ask any agent "show me how" followed by anything you want to visualize. Open Rebyte and give them a spin.