Restore the quarantined TUI e2e lane
The TUI lane was quarantined on 2026-07-21 while landing PR #2169. This is
unrelated to that PR's preview deploy and browser/Vitest parallelisation work.
apps/os/e2e/tui-test/run.ts is an explicit logged skip until this task is
complete; the specs remain beside it.
Evidence#
@microsoft/tui-test0.0.4 reuses a worker it has already terminated when a timed-out test retries, so the retry cannot be trusted.- The framework hardwires shared mutable state under the working directory and temporary directory. Concurrent workflow processes race on its transformed spec cache and zsh dotfiles.
- Making two workflows reliable in parallel required a package patch and a bespoke isolation/retry harness larger than the product surface currently justifies.
- The terminal UI also has known product bugs and currently has no users. Its headless shared-data-layer smoke remains available independently.
Work#
- Reduce each framework defect to a minimal upstream reproduction and check whether a newer release fixes it; otherwise upstream focused fixes.
- Fix the known
iterate chatproduct failures exposed by the retained specs. - Restore a small runner that builds and exercises the published CLI, gives every workflow attempt a disposable project, and owns no hidden retry layer.
- Keep concurrent cases isolated without a repository-local package patch or a copied-spec harness.
Exit criteria#
- Remove the no-op skip from
apps/os/e2e/tui-test/run.ts. - Both retained workflows pass against a fresh preview on their first attempt.
- A retry-disabled repeated preview run shows zero TUI flakes, no shared-state races, and no unexplained errors.
docs/testing.mdagain describes the lane as active coverage.