[LuisAlbertoMK]

No workflow runs the suite today, so a change is green only on the machine that wrote it. This adds the three steps a contributor runs by hand:

• npx tsc --noEmit
• npm run build — the CLI and both vitest projects read from dist/, and build also asserts the viewer bundle and the 29 copied grammars
• npx vitest run --project engine --project ui — explicit, so the log names the project that failed instead of hiding a viewer regression inside an engine failure

Choices, all easy to change

• pull_request + workflow_dispatch, no push to main: it spends no runner time on merges, and a release is triggered by hand anyway. Adding push: branches: [main] is one line if you want post-merge coverage.
• Node 22, the floor the bundled runtime ships and the first line with node:sqlite, which the database layer requires. A 20/24 matrix is possible but roughly triples the run for coverage the suite does not currently vary on.
• Ubuntu only. The suite already has Windows-gated tests; a Windows matrix over ~700 files is slow and flaky. Linux is the platform the packaged runtime is least exercised on.
• cache: npm relies on the committed lockfile.
• The artifact upload on failure is there because a build failure can come from the viewer bundle or a missing grammar, not only from TypeScript, and dist/ is what says which.

Signal

These three steps already ran green on a Linux runner over a tree containing upstream/main at 428b3c48 — typecheck, build, full suite, 11m12s. So this lands with a known-good result rather than a red first run.

Happy to trim any of it, or split the workflow from the artifact step if you would rather review them separately.

[LuisAlbertoMK]

Closing this. #267 already settled it — a checked-in CI workflow is a maintainer call, and the answer was no. I should have read that before opening; sorry for the noise.