​

Chat E2E Testing Notes

​

Goal

​

What Was Manually Verified

• Opening /chat while signed in renders the chat page and chat history.
• Submitting a new message adds a history row immediately.
• When /api/claude sends session_id, the URL changes to /chat/:sessionId.
• The assistant response renders in the selected chat.
• Refreshing /chat/:sessionId restores the selected chat instead of showing the new-chat empty state.
• The shared history row had invalid nested <button> markup; this produced a React hydration warning and was fixed by making the row container a keyboard-accessible role="button" wrapper.
• Chat completion was refetching /api/chat/sessions multiple times; this was reduced to a single React Query invalidation for the chat sessions key.

​

Automated Coverage

npm run test:e2e -- e2e/tests/chat-routing-and-persistence.spec.ts --project=rich-user

• New chat optimistic history row appears before the mocked Claude stream resolves.
• The mocked session_id event moves the browser from /chat to /chat/:sessionId.
• The assistant response renders after the SSE stream completes.
• Chat sessions are invalidated exactly once after completion.
• Reloading /chat/:sessionId renders the persisted historical prompt and response.
• Switching from one chat to another fetches only the selected session detail, preventing the previous current-session refetch pattern.

​

How The Mock Works

• GET /api/chat/sessions returns an in-memory session list.
• GET /api/chat/sessions/:sessionId returns in-memory messages.
• POST /api/claude waits until the test releases it, then returns a custom SSE payload:

event: session_id
data: {"sessionId":"session_e2e_new","runId":"run_e2e"}

event: text
data: {"text":"Cash flow is money moving in and out of your accounts."}

event: done
data: {}

​

Recommended Scenarios To Add Next

• Running historical session: GET /api/chat/sessions/:id returns status="running" with partial response, and the chat body shows the generating indicator without falling back to a full skeleton.
• Failed session: status="failed" renders the prompt plus recoverable error copy.
• New Chat during active stream: verify the stale stream cannot write into the newly visible session.
• Mobile viewport: submit on a small viewport and verify tab/background-style stream recovery behavior as far as Playwright can simulate.
• Rename/delete: verify mixed running, completed, and failed rows still rename/delete all rows for a session.

​

Manual Smoke Checklist

1. Open /chat.
2. Submit a short prompt.
3. Confirm the history row appears immediately.
4. Confirm the URL changes to /chat/:sessionId.
5. Refresh that exact URL and confirm the selected chat returns.
6. Click a different history row and confirm the page changes to that chat without flashing back to the previous chat or a skeleton.
7. Start a chat, click New Chat, send another prompt, and confirm the sessions do not mix.
8. Watch the terminal for repeated /api/chat/sessions/:id or excessive /api/chat/sessions calls.