real-imageslisted
Install: claude install-skill Tgaurav1k/real-images
# Real Images
Coding agents recreate a site's layout well but leave the images remote: either
the original CDN URL, or a stock photo from unsplash/picsum. Both break the
moment the user goes offline or the host blocks hotlinking, and stock photos are
simply the wrong pictures.
This skill fixes that **after** generation, without touching the code:
| script | job |
|---|---|
| `scripts/localize.py` | downloads every remote asset the project references, stores it locally, dedupes it, rewrites the URLs |
| `scripts/grab.py` | renders the real site in Chromium, downloads the images the browser actually used, and swaps the placeholders for them in the right positions |
## Hard rule: only URLs change
These scripts edit nothing but the URL substring. Never follow up by
"improving" the component structure, HTML hierarchy, CSS layout, JS logic,
state or animations. If a picture needs a different position, change the one
`src`/`url()` string, nothing around it.
## Workflow
### Case A - the clone points at the original site's CDN
```bash
python scripts/localize.py ./my-clone --dry-run # list what would be downloaded
python scripts/localize.py ./my-clone # download + rewrite
```
Result: `public/assets/{images,icons,fonts,videos,audio,models}/…`, a
`assets-manifest.json` next to them, and every `src`, `srcset`, `data-src`,
`poster`, `background-image: url(…)`, `@font-face src`, `<source>`, `og:image`
now pointing at `/assets/…`.
Falls back to `<project>/assets` with