The Rebuild Package
What's in the zip, file by file, and how REBUILD.md is organized.
The Rebuild Package is the main thing Mateo makes. It's a zip your coding agent can build from, prepared automatically after every capture. Export it from the Export menu on a capture.
Use the Rebuild Package only for sites you own or have permission to reproduce (see the Terms of Service, sections 4 and 5). To borrow the look of a site you don't own, use DESIGN.md.
What's in the zip
The zip unpacks to a folder named <domain>-rebuild/.
| File | What it is |
|---|---|
REBUILD.md | The build specification. Your agent reads this first. |
evidence.json | Every measured value from the capture. When anything is ambiguous, it wins. |
screenshots/ | Visual ground truth: desktop at five scroll depths, plus mobile. |
assets/ | The site's original downloadable assets: images, logos, icons, inline SVGs. |
DESIGN.md | The design language reference (supporting). |
AGENT_PROMPT.md | The ready-to-paste agent prompt. |
How REBUILD.md is organized
- How to use this package. How the files fit together and which one wins when they disagree.
- Build target & signature moments. What the page is and the moments that make it recognizable.
- Design tokens (measured). Colors, gradients, typography, spacing and component values from the live page.
- Page anatomy & geometry (measured at 1440×900). Where every section sits and how big it is.
- Assets & placement. Which files in
assets/go where. - Motion & interaction. Animations, scroll effects, and the hover and focus states Mateo observed.
- Responsive behavior. How the page changes between desktop and mobile.
- Unknowns & decisions. What the capture could not confirm. Nothing here is a requirement; your agent should ask you.
- Acceptance criteria. What a faithful build has to get right.
- Self-validation loop. Step-by-step comparisons against the screenshots. The build is not done until this passes.
Measured and written parts
The measured parts of REBUILD.md (tokens, geometry, assets) are assembled directly from the capture. An AI model writes the narrative parts (signature moments, composition, motion choreography, responsive notes and acceptance criteria) from the captured evidence.
Values are labeled by where they came from: measured, observed, derived, inferred or unknown. Anything inferred or unknown is something your agent should check with you instead of guessing.