Desktop Screenshots

The Desktop screenshots in Projects and Workspaces come from the real renderer, driven by WebdriverIO against the fixture app with mocked native commands. They are not mockups, and they are not hand-cropped.

Where the assets and the capture live

• Captured PNGs, served at /screenshots/*.png, 2× for retina displays and palette-compressed: docs/public/screenshots/desktop-{projects,run-composer,run-workspace}.png.
• Capture spec: apps/agent-desktop-e2e/src/docs-capture.spec.ts. It mocks the native commands with documentation-friendly data (research-bot, the "Research workspace" project, a "Laptop" location at /Users/you/research), then walks Projects, the composer, and a finished run's detail.
• Config: apps/agent-desktop-e2e/wdio.docs.conf.ts, which sets the window size and --force-device-scale-factor=2, and writes into docs/public/screenshots.

The capture is a separate Nx configuration, so the normal journeys never overwrite the committed images.

Regenerating

No Docker stack or daemon is needed: the fixture app serves the renderer and the native commands are mocked.

pnpm exec nx run @moltnet/agent-desktop-e2e:e2e --configuration=docs-capture

Compress before committing, as the landing screenshots do. Raw captures are around 130 KB each and compress to roughly a third with no visible change on this dark UI:

cd docs/public/screenshots
for f in *.png; do pngquant --quality=65-90 --speed 1 --force --output "$f" "$f"; done

Review the three PNGs before committing them. Timestamps in the run detail are relative ("2 hours ago"), so they change on every capture; that is expected.

When to regenerate

Regenerate when the Projects screen, the composer's effective run settings, or the captured-workspace panel changes shape or wording. Update the alt text in the guide in the same change: it describes what each screenshot shows, and it is what screen-reader users get instead of the image.