Install the screencli GitHub App and every pull request on a watched repo gets an AI-driven browser test against its preview deployment — a pass / fail / inconclusive verdict and a recording, posted as a PR comment before you start your review. This page walks the full path from connecting the App to reading your first verdict.
Prerequisites
- A screencli account. Sign in at screencli.sh/login.
- A repo that produces preview deployments. screencli tests the preview URL that Vercel, Netlify, Cloudflare Pages (or similar) surface as a GitHub deployment on the PR. If your PRs don't create a preview, there's nothing for screencli to open.
- A Pro plan to connect a new repo. Installing the App on a repo for the first time is a Pro action — free accounts hit the paywall at the connect step. Managing an already-connected install isn't gated. See Plan & Billing.
1. Connect GitHub
- Go to Settings → Integrations (screencli.sh/settings/integrations).
- Click Connect GitHub. You're sent to GitHub to install the screencli App.
- Choose All repositories or Only select repositories, pick the repos you want watched, and approve the install.
- You're returned to the Integrations page with a confirmation: "GitHub connected. Open a PR on a connected repo — screencli will run AI-generated browser tests and post the verdict as a comment."
Each connected account (personal or org) shows up as an installation card with its repository selection and install date. To change which repos are watched later, use Connect GitHub again (it becomes Manage once installed) — it sends you to the App's install settings on GitHub.
2. Configure a repo (optional)
Defaults work out of the box: once connected, a repo is enabled and screencli tests the deployment URL GitHub reports. Open per-repo settings only when you need to override that or your app needs a login:
- Baseline tests — smoke tests that run on every PR, alongside the PR-derived one.
- Preview URL pattern — point screencli at a different URL than the GitHub deployment.
- Preview deployment bypass token — get past a protected (password-gated) preview.
- Authentication — sign the agent in before it tests a logged-in flow.
To reach these, click Configure repos → on the installation card, then pick a repository. See Configure a repo for the full field-by-field reference and examples.
3. Open a pull request
Open (or update) a PR on a watched repo and let its preview deploy. When the preview is live, screencli:
- Reads the PR diff and the relevant source, and decides which user-facing behaviors the change affects.
- Writes and runs a single browser test that walks through and asserts them, recording the session.
- Posts a PR comment with the overall verdict and per-check results, and a recording thumbnail that links to the watch page.
A run is keyed to the commit, so screencli tests each head SHA once — pushing a new commit triggers a fresh run.
What you get
- A PR comment with the verdict —
pass(all checks succeeded),fail(a check failed), orinconclusive(the run couldn't reach a verdict, e.g. the preview timed out) — a one-to-three-sentence reason, and a thumbnail linking to the recording. - A shareable recording at
screencli.sh/v/<id>— the video of the agent driving your app. - Your dashboard — every run's status, verdict, and recording in one place.
Notes & limits
- Preview, never production. screencli runs against the preview deployment from your GitHub deployments; your production app is never touched.
- Disabled repos are ignored. With Run on every PR off, deployment events for that repo are skipped.
- One run per commit. Runs are idempotent per
(repo, commit); re-testing the same commit requires a fresh run. - Verdict aggregation. With baseline tests plus the PR-derived test, any failing check makes the PR verdict
fail; any inconclusive (and no failures) makes itinconclusive; all passing makes itpass. - Bring your own Anthropic key for unlimited runs — add it under Settings → Anthropic API key and your credit meter shows ∞ instead of counting down.