Orbit currently has no automated cross-platform deployment testing.
A renderer-init bug like the recent DOMPurify regression (#36) would
ship to whichever platform we didn't manually launch on. Wire up a
GitHub Actions matrix so every push gets validated on all three OSes.

Scope

Phase 1 — matrix build (small, high-value):

• .github/workflows/build.yml with runs-on: [macos-latest, windows-latest, ubuntu-latest]
• On each: npm install, cd app && npm install, npm run typecheck, cd app && npx tsc --noEmit, npm test, cd app && npx electron-forge package
• Verifies build succeeds and packages on every supported OS

Phase 2 — Electron smoke test:

• Playwright's Electron driver: _electron.launch() against the packaged app (or app/ dev mode with xvfb-run on Linux)
• Assert: renderer loads, #agent-status updates from initial state, no uncaught errors in renderer console
• Catches today's-class bugs (module-init throws killing the renderer)

Out of scope for now:

• Full UI interaction testing across OSes — GitHub macOS/Windows runners have weak display-server support, expensive to maintain. Manual QA + Phase 2 smoke covers ~80% of regressions.

Logistics

• Land on `main`, not on `notebook-rewire-pure-md` — every future branch should inherit CI.
• Once `notebook-rewire-pure-md` is merged, rebase/merge `main` back so it benefits.

References

• Playwright Electron API: https://playwright.dev/docs/api/class-electron
• electron-forge package: already wired in `app/package.json`