The dashboard

The dashboard is where you record on demand and browse everything you've captured. Unlike the GitHub App (which runs automatically on every PR) and the CLI (which runs from your terminal), the dashboard lets you kick off a cloud recording from the browser — no repo, no install, no CI. Type a URL and a prompt, and screencli spins up a cloud browser, drives it, and posts the recording back.

Prerequisites

A screencli account. Sign in at screencli.sh/login.
A Pro plan. Cloud recording from the dashboard is a Pro feature — free accounts hit the paywall when they press Start Recording. See Plans, credits & billing. (You can also bring your own Anthropic key to skip the per-run credit meter; sandbox compute still needs Pro.)

Record from the browser

The composer sits at the top of the dashboard.

Enter the URL to record — any http(s) address the cloud browser can reach (e.g. https://staging.example.com).
Describe what to record in the prompt box. Be concrete about the flow and any assertion, for example:
```
Sign in with demo@example.com and confirm the dashboard loads with no console errors
```
Press Start Recording (or ⌘/Ctrl + Enter). screencli dispatches a cloud sandbox and takes you to the recording's page, where the status updates live.

Both fields are required. If you're not sure what to type, use one of the Try: suggestion chips under the composer to pre-fill a URL and prompt. Your prompt is the assertion screencli checks — see Writing effective prompts for how to phrase it so you get a real pass/fail.

The Recon toggle

Turn on Recon to run an unrecorded scouting pass first: the agent explores the app to map it, then the main (recorded) run heads straight for the goal. It trades a bit of extra time for a more direct final recording. Recon is experimental — leave it off for simple flows.

Watch a run happen live

After you start a recording, you land on its detail page. While the run is in flight, the video panel shows a live agent feed — each browser action (navigate, click, type, wait, …) appears as the agent performs it, with spoken narration lines called out, and a step counter in the corner. The page polls every few seconds, so a run moves through its phases without a manual refresh:

recording — the agent is driving the browser.
processing — the run finished; the video is being composed.
uploading — the composed video is being pushed to storage.
ready — the video is available to play, download, and share.
failed — the run couldn't complete (the reason is shown on the page).

When the run reaches ready, the live feed is replaced by the playable video.

Your recordings

Below the composer, the Recordings grid shows every recording visible to you — your own, plus any created by teammates who share your workspace. Each card shows a thumbnail, the title (or prompt/URL), duration, view count, and date. Click a card to open its detail page.

Each card carries a status or verdict pill:

A verdict pill — ✓ pass, ✗ fail, or ⚠ inconclusive — appears when the run was a test (the agent was asked to assert something). See Verdicts.
Otherwise a status pill shows running while a run is in flight, or ready when it's done.

Hover a card and click the trash icon to delete a recording (removes both the video and its record). An unlisted tag appears on cards whose visibility isn't public.

Notes & limits

Cloud recording is Pro-only. Free accounts can sign in and view existing recordings, but pressing Start Recording opens the upgrade paywall.
Composer runs skip GitHub. They test the URL you give directly — no repo is cloned and no PR comment is posted. For automatic per-PR testing, install the GitHub App.
The default viewport is 1280×800. Composer runs record at this size.
Live status is polled, not pushed. The dashboard and detail page refresh every ~3 seconds while anything is in flight, then stop.