fix(desktop): revive Desktop Smoke E2E shard 4 after the outcome-first Projects redesign

[devin-ai-integration]

Summary

Desktop Smoke E2E (4) has been cancelled at the 30m timeout on every main run since #95. Root cause confirmed by local repro, not inferred: #95 (b57d26def) deliberately moved WorkspaceTabs under <details data-testid="project-plumbing">, closed by default (ProjectOutcomeDetail.tsx:227), and replaced the old overview panel with ProjectOutcomeLanding (ProjectsView.tsx:822). ~19 upstream project-* tests still drive the pre-#95 contract, so each died on getByRole("tab", …) at a 30s timeout; ×3 retries ≈ 95s/test consumed the whole job budget.

Per the founder decision on #109, the upstream specs are adapted (not skipped with Crew-native replacements per the #65 precedent). Recorded as D-034, matching the orchestrator's allocation.

1. Test adaptation — the fork delta

One helper, desktop/tests/helpers/projectPlumbing.ts:

expandProjectPlumbing(page) // waits for project-plumbing, opens it if `open` is absent, waits for tabs to be actionable

called at each project/repository-open site in project-inbox, project-issue-comments, project-pr-review, project-commit-detail. Idempotent, no assertion rewrites. The helper carries a sync-hint comment: on upstream refreshes of these specs, keep both sides and re-add the helper calls. No changes to e2eBridge.ts, so no collision with the #125 / #110 harness work.

Also in the spec layer: the two seeded events below now carry the repo's standard "mocksig".repeat(20).slice(0, 128); the issue-comments test synchronises on the "Comment posted." toast between submissions (pre-existing product behaviour the test simply never reached before); the overview-CSS test is retargeted from the deleted projects-overview-panel/projects-overview-stat nodes to projects-outcome-landing + project-outcome-card-*, preserving its intent (no background painted behind the cards) rather than being skipped.

2. Product fix — the outcome landing lost its work-items error state (regression from #95)

ProjectsWorkItemsLoadNotice was rendered by the pre-#95 activity feed. #95's filter === "all" branch passes only onOpen/projects/pullRequests, while the PR and Issues surfaces kept their error/retry wiring (ProjectsView.tsx:845–882). A failed work-items query therefore rendered a silently empty landing with no way to retry. The notice is restored above the landing with the same inputs. project-pr-review.spec.ts:954 now passes on merit — that test was not modified.

3. Product fix — one malformed relay event could kill the app's relay transport

User-facing bug, found while chasing the last red test, well beyond a CI lane. An event without a sig reaching isVerifiedRelayEvent() threw on event.sig.startsWith(...). The throw escapes handleWsMessage(), whose catch calls resetConnection() — which nulls wsId and rejects in-flight history subscriptions. Every subsequent relay operation then fails at RelayTransportLeaseAuthority.capture() with Relay is not connected., with no user-visible cause and no reconnect. Captured sequence from instrumented run:

[relay-debug] message handler error
  { generation: 3, error: "TypeError: Cannot read properties of undefined (reading 'startsWith')",
    message: ["EVENT","history-5d4f2ba3-…",{"kind":30617,…}] }   ← event has no `sig`
[relay-debug] resetConnection
  { message: "Cannot read properties of undefined (reading 'startsWith')",
    wsId: 2, generation: 3, subscriptions: [ …92 subscriptions… ] }
→ later mutation: fetchEvents() → capture() → "Relay is not connected."

That is 92 live subscriptions torn down by a single malformed event. isVerifiedRelayEvent() now fails closed on a missing/non-string sig or malformed tags instead of throwing, with unit coverage in relayEventVerification.test.mjs. Verification semantics for well-formed events are unchanged.

Related issue

Fixes #109.

Out of scope, per founder: project-outcomes.spec.ts:108 (Crew-only) still fails on getByTestId('message-thread-panel') — untouched here, owned by #125 (root-caused there as a spec-side race where window.__BUZZ_E2E_SEED_MOCK_MESSAGE__?.(…) runs before the bridge globals exist and silently no-ops).

Newly-visible preexisting failures surfaced by reviving the lane are tracked in #132 (see below).

Testing

Acceptance met on this PR's CI — Desktop Smoke E2E (4) completed instead of being cancelled:

Running 250 tests using 1 worker, shard 4 of 4
227 passed / 17 failed / 4 skipped — 22.1m

vs main run 31362178966, where shard 4 was cancelled at the 30m limit with no verdict.

Local, xvfb-run -a pnpm exec playwright test --project=smoke over the four affected spec files:

	before	after
result	~19 tests × 30s first-assertion timeout (31.4–31.6s in CI, ×3 retries)	32 passed, 0 failed, 4 skipped
wall time	shard killed at the 30m job timeout after test 106/250	1.9m

Reproduced the original failure first to confirm the diagnosis:

Error: locator.click: Test timeout of 30000ms exceeded.
Call log: - waiting for getByRole('tab', { name: 'Pull Request' })
  > 30 |   await page.getByRole("tab", { name: "Pull Request" }).click();

Also confirmed the clobber-guard test passes at b57d26def^ in a separate worktree, isolating the regression rather than assuming it.

Gates: pnpm --filter buzz build:e2e, pnpm --filter buzz check, pnpm --filter buzz typecheck, just test-unit (8 groups, 0 failed) all pass.

Docs: STATE.md records the revived shard-4 lane; DECISIONS.md adds D-034.

Remaining red checks are preexisting

• Shards 1 and 3 reproduce the identical failure sets from main run 31362178966 — shard 1 the same 8 (including channels.spec.ts:500, issue Agent question card never renders in the channel: channels.spec.ts:500 hard-fails on every main run since #96 #110), shard 3 the same 4 (inbox-edit.spec.ts:175/:325, inbox-reactions.spec.ts:36, messaging.spec.ts:1819).
• Shard 4's 17 failures could not be compared against a main verdict — that verdict has not existed since feat(desktop): outcome-first Projects with in-place thread panel #95, which is the whole point of this issue. So each was re-run locally under xvfb-run in a separate worktree at origin/main (35af74019) and on this branch: 16 of 17 fail identically on both, and overscroll-boundary.spec.ts:34 is flaky (failed once in CI, passed on retry, passes locally on main). relay-reconnect.spec.ts:617 was checked specifically because the verifier hardening touches relay behaviour — it fails identically on main, so it is not ours. Full list and grouping in Shard-4 Desktop Smoke E2E: 17 preexisting failures newly visible after the lane was revived (15 share a missing message-thread-summary) #132; 15 of them share a missing message-thread-summary and look like one shared cause.

Link to Devin session: https://app.devin.ai/sessions/ce99f3d1cda240b78a31902e1b527e3f
Requested by: @oscarlehuu

---

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 1 additional finding.