capture-store

---
name: capture-store
description: Retake the Chrome Web Store and Edge Add-ons screenshots in store/screenshots/, or the Mac App Store ones in store/safari/, with `yarn capture:store`, then check each one by eye before committing. Use when the UI has changed and the listing's screenshots are out of date, before a release, or when the user asks to "update the store screenshots", "retake the store shots", or change what one of them shows.
---

# Capture the store screenshots

`yarn capture:store` builds the extension, then takes each listing screenshot in a fresh
headless Chrome on the live site, saving 1280×800 PNGs into `store/screenshots/`. Chrome
takes the first five, Edge all of them. `yarn capture:store --help` lists them.

The script gets every shot to the same state each time; whether the result is a good store
image still takes a look, because the sites it shows change under it.

## 1. Run it

```
yarn capture:store
```

- `--only 7-readability 2` retakes some shots, by name or number; `--no-build` reuses `dist/`.
- `--safari` takes the Mac App Store set into `store/safari/` instead: every shot but the CLI, which Safari builds don't carry, at 2560×1600. Safari can't be automated with an extension loaded, so these are rendered in headless Chrome too; no browser UI shows, and the panel looks the same.
- `--phone` cuts the same screens out for the feature tour on stylebot.dev's phone layout, into `captures/tour/` (or `--out <dir>`) as WebP, for the site's repo's `src/assets/tour/`: the panel at Chrome's narrowest side panel width plus a strip of the page, with no headline (the site shows each as translated text, in `home.tour`). Each shot's `phone` entry sets its layout; check that panel text still reads at a phone's card width.
- It takes about a minute. A shot that fails says why and the others still run.
- The command line shot drives the real CLI in a temporary HOME, as its e2e spec does, so
  it never touches `~/.stylebot` or the user's browsers.

## 2. Look at every image

Open each PNG and check, before saying they're done:

- **The site loaded.** A bot check, consent dialog, or error page instead of the site means
  that site now blocks headless Chrome (NYTimes does): pick another and say so.
- **Nothing reads as broken.** A warning icon in the panel, a wrapped or clipped row, text
  justified into big gaps, a menu cut off at the edge.
- **The panel is in the theme it should be.** Every shot's Stylebot panel is dark; pages
  alternate between dark and light so the set doesn't read as a dark-mode extension.
- **The slide's crop shows what its headline is about.** Every store gets slides: a
  headline over an enlarged crop of the screen, so the UI reads at the size
  the store shows its images. Nothing that matters should be cut at the window's edges.
- **The terminal's last line still describes the page beside it.** It's written by hand
  (`SUMMARY` in `cli-shot.mjs`); rewrite it whenever the Newspaper theme changes.

Send the user the images in order, then commit them once they're happy.

## Changing a shot

Each shot is an entry in `scripts/capture/shots.mjs`: the site, the theme it's seeded
with, what's opened or hovered, and its `slide` — the headline (`*word*` is highlighted),
accent, and either a narrower `pageWidth` to lay the page out at or a `crop`.
Keep the copy free of product names other than Stylebot's. Gallery themes are read from the site repo's gallery, `src/assets/gallery/<site>/` in a
stylebot-site checkout beside this one, including the Hacker News profiles the gallery also
lists; Modern, which it doesn't, is kept in `scripts/capture/hn-modern.css`. When the user
updates one in their browser, refresh its file with `yarn stylebot css get news.ycombinator.com --profile "<name>"`, keeping the
gallery file's `/* Site: Name */` first line, then retake the shots and, with the `capture-gallery` skill, the gallery image.

Choosing what to show:

- **Pick an article whose opening has no images** for Wikipedia in Slate: its two justified
  columns squeeze around pictures and leave gaps between words.
- **Check a site loads headless** before moving a shot to it. Reddit, Stack Overflow and
  NYTimes block it; Gmail needs a sign-in.

More agent context in ankit/stylebot

7 other files this repository gives its agents.