Trigger self-hosted PR checks on demand with @openreview

Who it's for

This skill is for people who want to host their own AI code-review bot and can deploy it to Vercel. Officially it is open source and self-hostable: after you deploy it to Vercel and connect a GitHub App, you can run Claude-powered checks on a Pull Request on demand. It also fits people who already have a test PR and are willing to mention the bot in a comment.

It is not for people who cannot deploy or connect a GitHub App and treat “one local command” as the starting point. The source does not provide a local installer. It is currently in beta. It began as an internal project the Vercel team used to try their own technology together, so behavior may be rough.

The official repository is https://github.com/vercel-labs/openreview Follow the path already written on that page. The runtime is described as an isolated Vercel Sandbox; see https://vercel.com/docs/sandbox Do not picture it as running the full repository directly on your laptop.

First try in 3 minutes

Follow only the one path the official docs write: deploy to Vercel and connect the GitHub App first, then mention @openreview in a PR comment.

Prerequisites: you have permission to deploy to Vercel, and a repository where you can install a GitHub App. Confirm first that both steps will change remote configuration. If you do not want to deploy or grant access yet, only read the official docs and do not run a deploy.

Where to install: follow the official page to deploy the app to Vercel and connect the GitHub App. Do not invent install commands that do not appear there.

Low-risk task: pick a non-production Pull Request that can be discussed in public. Before you write @openreview in a comment, confirm that this comment will trigger a remote check. Do not run the first trial on a release branch or a PR that contains secrets. After you trigger it, only watch whether the bot replies; do not let it change your local files.

Success signal: that PR shows a check response triggered by @openreview. If there is no reply, first verify that the app is deployed and the GitHub App is connected; do not switch to an installer that was never listed. The official docs also say it is powered by the Chat SDK; the page is https://chat-sdk.dev/ Use that only for comparison — you do not need to change that layer on the first try.

Use cases

Case 1: Trigger one on-demand check on any PR. The source says mentioning @openreview in any PR comment is enough to trigger it. This fits people whose repository is already connected and who only want feedback on one change. The comment will be publicly visible, so confirm before you send it.

Case 2: Add custom skills for the bot. The source says it ships with review skills and supports custom skills through .agents/skills/; built-in skill docs point to https://skills.sh/ This fits people who can already trigger @openreview and want to put extra skill files in that directory. Confirm that writing into .agents/skills/ is a file change before you do it; the source does not list every field of the skill file format, so do not invent a spec.

Limits

OpenReview is currently in beta. It is described as open source and self-hosted. The main path is deploy to Vercel, connect a GitHub App, then trigger with @openreview. Do not put an installer that never appears into the main path.

Work runs in an isolated Vercel Sandbox. The source does not say, in any sentence you can see, which local commands it will run, so do not assume it will always run tests or formatters. The only custom skill directory written out is .agents/skills/.

The official page is https://github.com/vercel-labs/openreview If an agent suggests a trigger other than @openreview, or replaces the page above with an address that never appears, treat that page as the source of truth.