I kept hitting the same failure in Cursor: drop a real HTML page (or a PDF) into the agent and ask for a redesign. It invents a vibe. Navy glass, a fake dashboard. The nouns already in the file never make it onto the page.
So I wrote a skill that has to extract *this* source’s nouns, dates, and colors and ship from those. A Texas notebook gets Lone Star / Alamo / Rio Grande. A Jules Ice Cream source gets parlor / cone / freezer / flavor board. Same tokens, not a reskin. If a live run could be mistaken for the Texas gold poster when the source isn’t that notebook, that’s a fail.
Forms that actually do different jobs:
- **infographic** — paper poster of facts already in the file (not an ops dashboard)
- **svg** — motion lives on the drawing
- **3js** — a room you can orbit
- **simulation** — a playable model of facts already named
- **webpage** plus a register word (cinematic, landing, artistic)
Install in Cursor:
```
npx skills add Kayforkind/reimagine-it
```
Then `/reimagine-it infographic` (or `3js`, `svg`, `simulation`) on a real file.
No-install gallery: reimagine-it — redesign the file you actually have
MIT. Not a paid seat. Built with Cursor. Looking for people to run it on their own HTML and tell me where it still mood-boards.