frontend-dev Handoff — E2E Helper Fix — 2026-04-22

Branch: feat/backend-migration-e2e-skill-library Tip SHA: aa8042fa3 (docs(e2e): note aionui-backend must be on PATH for tests) Predecessor report: docs/backend-migration/e2e-reports/2026-04-22-skill-library.md (0 PASS / 29 FAIL at goToSkillsHub timeout)

Summary

The Skill-Library e2e suite was 29/29 FAIL with every test blocked in its beforeEach at [data-testid="my-skills-section"] visibility. Two layered causes were in play — a stale renderer bundle and legacy-IPC helpers — plus a latent PATH issue that was masking backend connectivity. All three are now resolved and a representative test (TC-S-01) passes in ~13.6s on this branch.

Commits (atomic)

All commits pushed to origin.

Root causes (evidence-based, per Playwright trace + screenshot)

1. Stale renderer bundle in out/renderer/

Trace stack (.../out/renderer/assets/index-DQXhCJEQ.js) revealed Electron was loading out/renderer/index.html from 2026-04-21 16:52 — before the HTTP-bridge migration landed. bun run test:e2e does not build; it loads whatever bundle is on disk (per tests/e2e/README.md §1). The stale bundle still used legacy IPC, never set window.__backendPort, and rendered SkillsHub in an older layout where the my-skills-section element either didn't exist or mounted differently.

Fix: bunx electron-vite build before running e2e. Documented in the README update at aa8042fa3.

2. Legacy IPC in tests/e2e/helpers/skillsHub.ts

Commit 5c4b010f5 deleted src/process/bridge/fsBridge.ts and its subscribe-<key> handlers. The helpers still emitted: list-available-skills, list-builtin-auto-skills, get-skill-paths, detect-and-count-external-skills, import-skill-with-symlink, delete-skill, add-custom-external-path, remove-custom-external-path, get-custom-external-paths. Each emit hit the 10s IPC timeout with no receiver.

goToSkillsHub's health check probed get-skill-paths up to 5 × 3s = 15s of wasted time per test, compounding the proximate timeout.

Fix (commit 000676801):

• Added tests/e2e/helpers/httpBridge.ts — a thin page.evaluate-based client that hits http://127.0.0.1:<window.__backendPort>/api/skills/* and unwraps the backend's { success, data, ... } envelope, mirroring src/common/adapter/httpBridge.ts:76.

• Rewrote every skills helper against the HTTP routes defined in src/common/adapter/ipcBridge.ts:268-363:

  Legacy key	HTTP route
  list-available-skills	GET /api/skills
  list-builtin-auto-skills	GET /api/skills/builtin-auto
  get-skill-paths	GET /api/skills/paths
  detect-and-count-external-skills	GET /api/skills/detect-external
  import-skill-with-symlink	POST /api/skills/import-symlink
  delete-skill	DELETE /api/skills/{skillName}
  add-custom-external-path	POST /api/skills/external-paths
  remove-custom-external-path	DELETE /api/skills/external-paths?path=...
  get-custom-external-paths	GET /api/skills/external-paths

• Preserved the { success: boolean; msg?: string } return shape on mutation helpers so existing expect(importResult.success).toBe(true) assertions in test bodies keep working.

• Bumped my-skills-section visibility timeout from 5s → 15s to accommodate dev-mode cold boot (Vite dev + Arco Tabs + initial list render takes 6-10s after navigation).

• Fixed a bug in tests/e2e/features/settings/skills/edge-cases.e2e.ts: the afterEach invoked a non-existent IPC key add-external-skill-source (typo — it was never a real handler). Replaced with the canonical addCustomExternalPath helper.

• Re-exported invokeBridge from skillsHub.ts for backwards-compat so tests/e2e/helpers/index.ts:50 continues to resolve.

3. aionui-backend binary not on PATH during tests

Step A's first rerun against the rebuilt bundle still hit Failed to fetch (127.0.0.1:13400). Diagnosis: window.__backendPort was 0. The Electron main process calls backendManager.start() at src/index.ts:467, which calls resolveBinaryPath() at src/process/backend/binaryResolver.ts:23 — that tries which aionui-backend. My shell did not have ~/.cargo/bin on PATH, so the resolver returned nothing, backend never started, port stayed at 0, and every HTTP call fell through to the hardcoded fallback 13400 — which nothing was listening on.

Fix: documented in tests/e2e/README.md §2. Operationally: export PATH="$HOME/.cargo/bin:$PATH" before bun run test:e2e. This is what the predecessor e2e-tester was implicitly doing (their shell had it).

Step A trace evidence

Captured via E2E_TRACE=1 bun run test:e2e tests/e2e/features/settings/skills/core-ui.e2e.ts --grep TC-S-01.

• Trace zip: tests/e2e/results/features-settings-skills-c-4f679-h-builtin-and-custom-skills/trace.zip (not committed — transient).
• The failure screenshot showed my-skills-section DID render eventually ("我的技能 25" header visible at screenshot time t=20.5s), confirming the element was structurally fine in the rebuilt bundle.

Step D verification

PATH="$HOME/.cargo/bin:$PATH" E2E_DEV=1 bun run test:e2e \
  tests/e2e/features/settings/skills/core-ui.e2e.ts --grep "TC-S-01" \
  --reporter=list
# ✓ 1 TC-S-01: should render My Skills section with builtin and custom skills (13.6s)
# 1 passed

A second single-test spot check (TC-S-06) revealed a test-data issue unrelated to the helpers: the test asserts at least one builtin skill exists in the fresh sandbox, which is environment-dependent. Routing that to e2e-tester-2 per task scope — this handoff does NOT attempt to fix test-body assertions.

Known issues left for e2e-tester-2

1. Full 29-test suite not run here. Task scope explicitly reserves that for e2e-tester-2 ("do NOT invoke e2e suite yourself; e2e-tester-2 retests"). Any downstream failures beyond goToSkillsHub are in the test bodies and predate this task.

2. Pre-existing lint no-await-in-loop errors in cleanupTestSkills (lines 483, 495). Not touched — pre-existed before my migration. Can be fixed with Promise.allSettled pass when someone is doing lint-cleanup work.

3. invokeBridge still imported from bridge.ts by non-skills helpers (extensions.ts, chatGemini.ts, chatAionrs.ts) — all out of scope for this task. Those will need the same migration in future waves.

4. out/renderer/ staleness is a recurring footgun. Consider adding a mtime check at fixture startup (fixtures.ts) that warns if out/renderer/index.html is older than the most recent commit on src/renderer/. Defer.

Quality checks on exit

• bunx tsc --noEmit: clean (no new errors).
• bun run format over touched files: clean (oxfmt applied).
• bun run lint over touched files: 9 warnings, 1 error — all pre-existing in cleanupTestSkills (confirmed via git stash).
• Single-test verification on this branch: PASS (TC-S-01, 13.6s).

Pointer

Branch tip: aa8042fa3. Next action is e2e-tester-2 re-running the full 29-test suite.

Coordination notes

• e2e-tester-2 message sent at handoff with sha + one-line trace summary.
• Coordinator updated throughout (trace plan, trace findings, backend port discovery, final pass).