[quanru]

Add browser regression coverage for two core drawDB workflows: creating related tables and restoring them after a reload, and importing SQL, renaming a column, and exporting the updated schema.

Midscene operates the visible UI and checks the rendered diagram. Read-only IndexedDB checks verify the saved tables, fields, primary keys, and relationship endpoints; the export check parses the actual downloaded SQL. Each case starts with isolated browser storage, runs against a production build, and has no case retries.

The opt-in GitHub Actions workflow runs on trusted pushes or manual dispatch. It preserves screenshots and reports even when tests fail, optionally publishes the interactive report to Cloudflare Pages, and adds report/artifact links to the job summary. No model or Cloudflare credentials are required for the existing PR build workflow. Setup instructions are included in tests/midscene/README.md.

Validation:

• Local build, ESLint, TypeScript/node registration checks, and workflow validation passed.
• Both browser cases passed locally using OpenAI gpt-6.1-terra in 2m 46s.
• Fork CI passed on Ubuntu 24.04: both cases passed in 2m 37s with no case retries.
• CI summary and report link
• Public interactive report
• The upstream Node 20 and Node 22 build checks passed. The upstream Vercel preview requires maintainer authorization.

Model and Cloudflare credentials are configured only in quanru/drawdb; the public report can be opened without those credentials. Cloudflare publication and the clickable GitHub summary link were verified on the fork.

[vercel]

@quanru is attempting to deploy a commit to the dottle's projects Team on Vercel.

A member of the Team first needs to authorize it.