Summary

Running bun test browse/test/sidebar-ux.test.ts on the current main branch (a5833c41 — v1.57.10.0) produces 73 failures and 4 errors out of 176 tests.

Reproduction

git checkout main # a5833c41
bun test browse/test/sidebar-ux.test.ts
# → 103 pass, 73 fail, 4 errors

Confirmed on macOS (Apple Silicon). These failures exist on upstream/main before any branch changes are applied.

Failure clusters

1. pickSidebarModel routing

error: expect(received).toBeGreaterThan(expected)
Expected: > 0 Received: -1
at sidebar-ux.test.ts:1665 — routerFn.indexOf('ANALYSIS_WORDS')

The test extracts pickSidebarModel from server.ts and checks the source contains ANALYSIS_WORDS / ACTION_PATTERNS. Both are absent from the current output — indexOf returns -1.

2. LLM-based cleanup prompts — 4 tests fail

3. Sidebar arrow hint / tool_use collapse / idle timeout / shutdown — ~15 tests fail

Suspected cause

The Codex review default-on PR (#1966, v1.57.10.0) likely refactored or inlined ANALYSIS_WORDS / ACTION_PATTERNS in browse/src/server.ts. Tests at lines 1651–1690 still expect those symbol names to appear literally in the extracted function source.

Impact

• bun test gives a dirty run on main — blocks contributors from knowing if their PR introduced new failures
• Makes it hard to bisect regressions

Ask

1. Confirm if this is already tracked
2. If the refactor was intentional — update tests to match the new structure
3. If not — restore the constants or fix the extraction logic

Happy to send a patch once the fix direction is confirmed.