ci(electron): Test Electron Framework permanently red on main since #606 — stale chat-app assertions

[itomek]

Why this matters

The Test Electron Framework job has been red on every PR that triggers it since PR #606 merged. The most recent green run on main is from 2026-05-08 — every dependabot bump or contributor change under src/gaia/electron/**, src/gaia/apps/*/webui/**, src/gaia/agents/emr/dashboard/electron/**, or tests/electron/** now lands on red CI, forcing every reviewer to manually decide "is this red mine?" before approving. That's exactly the wrong default for a CI signal.

Surfaced while reviewing #1194 (Electron 35→42 EMR bump). The same two failures also appear on #1195, #1196, and #1199 — i.e. unrelated to any of those PRs.

Failing assertions

1. test_electron_chat_app.js — .empty-chat* selectors no longer exist

tests/electron/test_electron_chat_app.js:1176-1179:

expect(chatCss).toContain('.empty-chat');
expect(chatCss).toContain('.empty-chat-title');
expect(chatCss).toContain('.empty-chat-chip');

grep -n "empty-chat" src/gaia/apps/webui/src/components/ChatView.css returns nothing. The classes were removed when #606 reworked the empty-state UX.

2. test_electron_chat_installer.js — MemoryDashboard.tsx is 159 KB, breaches 100 KB cap

tests/electron/test_electron_chat_installer.js:397:

expect(stats.size).toBeLessThan(100 * 1024);

src/gaia/apps/webui/src/components/MemoryDashboard.tsx is 159,276 bytes — about 1.55× over the rule. Added wholesale by #606.

Root cause

Both violations were introduced in the same commit: 74f637a4 — feat(memory): agent memory v2 — second brain with hybrid search, LLM extraction, and observability dashboard (#606). The PR introduced new empty-state UX plus a large new dashboard component but did not update the corresponding tests.

Why it went unnoticed for ~2 weeks

.github/workflows/test_electron.yml has a paths: filter — it only runs when the diff touches the listed directories. No main commit between 2026-05-08 and 2026-05-21 has touched any of those paths, so the workflow simply hasn't run on main. The dependabot batch #1194–#1199 was the first thing to re-trigger it, which is when the red surfaced.

This is a latent CI-gap class: when a path-filtered workflow is broken on main, every contributor pays a "is this red mine?" tax until someone fixes it.

Acceptance criteria

• Test Electron Framework job is green on a fresh PR that touches one of the path-filter paths (verify on chore(deps): bump electron from 35.7.5 to 42.2.0 in /src/gaia/agents/emr/dashboard/electron in the emr-dashboard-dependencies group #1194 or chore(deps): bump the example-app-dependencies group in /src/gaia/apps/example/webui with 5 updates #1195 after the fix lands).
• test_electron_chat_app.js resolves the missing .empty-chat* assertion. Pick one:
  • A — update the test (preferred if feat(memory): agent memory v2 — second brain with hybrid search, LLM extraction, and observability dashboard #606's UX rework was intentional): assert selectors that actually exist in the current ChatView.css, chosen so the test still fails if someone removes the empty-state UX again.
  • B — restore the selectors to ChatView.css (only if their removal was unintentional in feat(memory): agent memory v2 — second brain with hybrid search, LLM extraction, and observability dashboard #606).
• test_electron_chat_installer.js no longer fails on MemoryDashboard.tsx. Pick one:
  • A — raise the cap to a meaningful new value (e.g. 200 KB) with a one-line comment explaining why and what triggered the change.
  • B — split MemoryDashboard.tsx into smaller files so the existing 100 KB rule holds.
  • C — allowlist specific known-large dashboard files and keep the 100 KB cap as the default for the rest.
• If choosing 1-A or 2-A above, the new assertions catch a meaningful regression — they should fail if the underlying UX/architecture rule is broken again, not just "string is present".
• No new Test Electron Framework failures introduced by the fix.

Test plan

• cd tests/electron && npm test -- test_electron_chat_app.js test_electron_chat_installer.js — both pass on the fix branch.
• Re-run Test Electron Framework on the fix PR; confirm green.
• Sanity-check the path filter still triggers: push an unrelated whitespace-only change to one of the filtered dirs on the fix branch and confirm the workflow runs.

Optional follow-up (separate issue if worth tracking)

Test Electron Runtime (example) and Test Electron Runtime (jira) jobs actually boot Electron under xvfb, but there's no equivalent for src/gaia/agents/emr/dashboard/electron/. That gap is why #1194 (a major-version Electron bump) had to be verified by hand on a Linux box — there's no CI runtime signal for that app. Adding the third runtime job would catch the next major Electron bump automatically.