test(desktop): follow split agent cards and trailing header actions

[mfethe1]

Summary

Two desktop smoke tests use assumptions that predate the current agent cards and channel header. On PR #21, the archived-agent test times out looking for an unsplit persona card even though the live sibling's named card is visible. The narrow-header test omits the gap before the trailing Tasks button and reports 58 pixels where it allows 52. Both failures repeat on every retry, and the desktop source tree is identical to the PR's base.

Select the surviving named agent card and assert that its profile is Sibling B, preserving the archived-key and lifecycle checks. Measure menu adjacency against the actual trailing Tasks control and its computed group gap; retain the non-overlap, edge-padding, and menu-action checks. Wait for existing layout animations before measuring. Only the two test files change.

Related issue

Observed in PR #21's agent-profile smoke job and header smoke job. Two other retry-sensitive failures in those jobs remain separate investigations; this patch does not claim to resolve them.

Testing

Candidate d55b27ef512c790487855342e85a0d0b054044fd, based on adeec6eb5c4ff425aa03d8a744a295da4429f67f.

• Actual E2E build and production Playwright smoke configuration on Airy, with retries disabled: the baseline reproduces both failures, with one other archived-agent case passing. The repaired and restored runs both pass all three affected cases.
• Three deliberate regressions fail: restoring the obsolete selector, shifting the Tasks control, and shifting the header group into the thread. All mutations were restored.
• Biome, TypeScript, and the diff whitespace check pass.
• Independent review of the complete exact diff is clean; the reviewer independently ran the actual affected Playwright flows and observed all three passing.
• Full local just ci after documented dependency setup: exit 1 at the final mobile stage, with 2,241 passed and eight baseline golden comparisons failing. Desktop JavaScript tests passed 6,583/6,583 and preceding recipe stages completed. All eight rendered mobile PNGs are byte-identical to the saved exact-base Airy outputs; expected images are unchanged.
• Hosted run 34150357391 passed all three changed browser flows on merge 95f230d87f53b72ed0eba6b0cff42233b9a31d67, whose parents match the exact base and head above. The overall run failed: another smoke shard had five failures, and two desktop integration jobs failed. Four smoke flakes and three skips also remain recorded. Those other failures are not classified as baseline without separate diagnosis.
• Full CI remains red, so this stays a draft and is not merge approval.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[mfethe1]

🧪 Jev PR-Triage Pilot

PR #26: test(desktop): follow split agent cards and trailing header actions — verdict: REVIEW

question	answer	confidence
sensitive_path	P(yes)=0.03	—
test_coverage	P(yes)=0.99	—
size_concern	P(yes)=0.09	—
risk	0.49/2	0.26
route	needs_review	0.76
block_reason	none	0.92

Block reason (if any): none
Pilot: Jev (System One) typed triage. Verdict thresholds in code; escalate on low confidence.

[mfethe1]

🧪 Jev PR-Triage (v3)

PR #26: test(desktop): follow split agent cards and trailing header actions — verdict: REVIEW

dimension	metric	value
Correctness	problem resolved	P(yes)=0.51
Correctness	rework needed	0.63/2
Quality	code quality	1.77/2
Quality	test coverage	P(yes)=0.97
Architecture	integration compatibility	1.55/2
Architecture	future proof	1.05/2
Governance	sensitive path	P(yes)=0.04
Governance	size concern	P(yes)=0.15
Synthesis	merge risk	0.48/2
Synthesis	route	needs_review (0.59)
Diff	security exposure	P(yes)=0.02
Diff	performance impact	0.00/2

🔍 Reviewer focus

• File: desktop/tests/e2e/messaging.spec.ts
• Concern: correctness_edge_cases

Block reason (if any): none
Pilot: Jev (System One) typed triage, 15 metrics + reviewer focus. Verdict thresholds in code.