Taking screenshots
Screenshots of the web app show a change to reviewers and illustrate the docs. Take them the same way each time, so a before and an after compare directly.
The standard capture
Section titled “The standard capture”Start the app, then capture it:
bun run dev:web # the public web (apps/web), running Mob in the browserbun scripts/screenshot.ts http://127.0.0.1:5174/app.html --name workspace --wait textareascripts/screenshot.ts writes <name>-<size>-<theme>.png for every size in
both themes, into $TMPDIR/mob-screenshots unless --out names a folder:
| Setting | Value |
|---|---|
| Sizes | desktop 1280×800, mobile 390×844 (--sizes) |
| Themes | light and dark, set through localStorage['starlight-theme'] before the page loads (--themes) |
| Density | deviceScaleFactor: 2 |
| Motion | reducedMotion: 'reduce', so the calves and transitions hold still |
| Ready | --wait <selector> waits for an element; otherwise the page settles for --delay ms (1500) |
| State | --storage key=value sets localStorage first, such as mob-browser-limits-seen=1 to skip the first-use notice |
| Page | the viewport; --full captures the whole scrolling page |
It uses Playwright with Chromium, which is not a dependency of the repository:
cloud sessions have both installed; elsewhere run bun add -g playwright and
bunx playwright install chromium.
Which app to capture
Section titled “Which app to capture”- The public web (
bun run dev:web, port 5174) runs Mob in the browser on the demo model, with no service to start. Use it for/app.html, the landing page (/landing.html),/demo.htmland/account.html. - A local Mob shows what needs a runner (agents, plugins, the sandbox):
run
./build/mob serve --demoandbun run dev, and sign in with the token it prints. Seed rich conversations withPOST /api/conversations/importrather than typing them.
States the script cannot reach
Section titled “States the script cannot reach”Dialogs, menus, approvals, errors and several tabs need a few steps first. Write a short Playwright script for them outside the repository, with the settings above, and:
- Click and type through roles and labels (
getByRole,getByLabel), as a user would. - Open several tabs as pages of one browser context: they share storage, Web Locks and the browser runner, like tabs of one browser.
- Make a slow reply with a local OpenAI-compatible server that streams slowly and allows CORS, saved as a custom provider.
- Crop to the part that changed with
clipwhen the whole window says less.
Sharing them
Section titled “Sharing them”- Put screenshots in the pull request, before and after, in both themes, and on mobile when the layout changes.
- Commit only images the docs use, under
docs/images/.