DEVOPS
AI visual judge that scores previews and pages on-call for breakage
Sends preview screenshots to a vision model that judges layout, broken images, and overflow.
How it runs
The automated pipeline, trigger to output.
- TriggerVercel preview deploy readyVercel
- ActionCapture route screenshotsBrowserbase
- ActionVision model scores layout and defectsOpenAI
- LogicBranch: score below pass threshold?
- ActionOpen PagerDuty incident with findingsPagerDuty
- OutputSet failing GitHub commit statusGitHub
What it does
Instead of relying only on pixel diffs, this workflow asks a vision model to act as a QA reviewer. It captures each critical route on the Vercel preview and prompts the model to flag broken layouts, missing or 404 images, text overflow, and obviously broken components, returning a structured score and an issue list. Builds below the quality bar are treated as production incidents.
When to use it
Use it when pixel diffing misses semantic breakage, like an image that loads as a gray box or a button that overflows its container on a fresh page that has no baseline. It suits teams who want a judgment call on net-new pages where a baseline comparison is impossible.
How it works
- 1A Vercel preview-ready webhook starts the run.
- 2A headless browser captures full-page screenshots of the configured routes.
- 3Each screenshot is sent to a vision model with a rubric to score quality and list defects.
- 4A branch checks whether the aggregate score is below the pass threshold.
- 5If it fails, a PagerDuty incident is opened with the offending pages and findings.
- 6A failing GitHub commit status is set so the deploy cannot be promoted.
Set it up
What you configure once, before turning it on.
- 1Connect VercelDeploys, runtime logs, analytics.
- 2Connect BrowserbaseHeadless browsers, sessions, replays.
- 3Connect OpenAIModels, embeddings, files.
- 4Connect PagerDutyIncidents, on-call, escalations.
- 5Connect GitHubRepos, issues, pull requests, actions.
- 6Set each agent's modelWe leave models unset so you pick the tier — fast + cheap, or top-quality.
- 7Tune it to your dataEdit the prompts, filters, and field mappings so it matches how your team works.
- 8Test, then turn it onRun once against a sample, confirm the output, then enable the trigger.
More DevOps workflows
Hugging Face Spaces idle-runtime sweep with auto-pause
On a schedule, scans all Hugging Face Spaces for ones running idle past a threshold, pauses them to stop billing, and posts a Slack summary with the estimated monthly savings.
Slack-approved pause for idle Hugging Face Spaces
On a daily scan it finds idle paid Spaces and posts an interactive Slack approval; on approve it pauses the Space and logs the decision to a GitHub issue audit trail.
Generate a weekly de-flake report and assign Linear cleanup tickets
On a weekly schedule, aggregates the current quarantine manifest and recent flake history, builds a prioritized report.
Block costly Hugging Face Space hardware upgrades in PR review
When a pull request changes a Space's hardware config, it estimates the new monthly cost and posts a GitHub PR comment that flags upgrades crossing a budget ceiling.
Auto-release tests from quarantine once they prove stable
Triggered by a webhook from a nightly stability runner, checks whether quarantined tests have passed enough consecutive runs, removes the stable ones from quarantine in GitHub.
Quarantine a test on demand from a PR comment command
Triggered when an engineer comments a quarantine command on a pull request, validates the test name, commits the quarantine change to that PR branch, opens a tracking issue.
Run it inside a business
This workflow drops into a full company template. Import the org, and this is one of the playbooks its agents run.
Run this workflow in your colony.
14-day trial. No DevOps. No Sales call. Provisioned in under a minute.