# preview_open reports "timed out after 660000ms" when the real failure is a 15s page.goto navigation timeout · Issue \#24278 · pingdotgg/t3code

[View on GitCafe](https://git.cafe/pingdotgg/t3code/issues/24278)

Repository: [pingdotgg/t3code](https://git.cafe/pingdotgg/t3code)

Visibility: public

State: completed

Priority: 0

## Description

[Discussion and activity](https://git.cafe/pingdotgg/t3code/issues/24278)

---

### What happened

An agent restarted a local Vite / TanStack Start dev server and called `preview_open({url: "http://localhost:1337/", reuseExistingTab: false})` about 25 seconds later. The call failed after **about 16 seconds** with:

```
Preview automation open timed out after 660000ms.
```

The server trace shows the real cause: `[cause]: Error: page.goto: Timeout 15000ms exceeded.` This reproduced twice, on two nightlies. Once the dev server was warm, the same call succeeded.

Because the message claims an 11-minute timeout, the agent concluded that preview was "stuck for 11 minutes" in this environment and started installing its own Playwright Chromium as a workaround.

### Diagnosis

1. On the server-browser host, the open request navigates with `waitUntil: "load"` and `timeout = Math.min(request.timeoutMs, NAVIGATION_TIMEOUT_MS)`, which is 15000 (`apps/server/src/preview/ServerBrowser.ts:1780`, `:1429`). A cold Vite dev server often needs more than 15 seconds to fire `load` on its first request.
2. The host replies with a `PreviewAutomationTimeoutError`. `classifyResponseError` (`apps/server/src/mcp/PreviewAutomationBroker.ts:210`) builds a new error from the broker's request context, whose `timeoutMs` is the 660_000 open budget (`:556`). The `message` getter (`packages/contracts/src/previewAutomation.ts:952`) prints that context value, and the remote message ("page.goto: Timeout 15000ms exceeded") is lost from what the agent sees.

Suggested directions:
- Report the host's actual timeout, or include its message.
- Consider letting `preview_open` accept a `timeoutMs`, or use a longer or `commit`-level wait for the initial navigation.

### Repro steps

1. Run T3 Code as a server (background service) with the server browser.
2. Start a Vite dev server whose first page load takes more than 15 seconds (cold dependency optimization or SSR compile).
3. Right away, have an agent call `preview_open` with that URL.
4. Observe `Preview automation open timed out after 660000ms.` returned in about 15–18 seconds.

### Environment

- Server: `0.1.3-nightly.20261023.3592` (also seen on `0.1.2-nightly.20261021.3473`), Linux x64 VPS, background service.
- Client: T3 Code Nightly desktop app on macOS, connected to the remote server over Tailscale.
- Provider: Claude agent (claude-opus-5-5).
- Triage CLI: 0.1.2.

### Evidence

- `PreviewAutomationBroker.invoke`: started 23:33:52.197Z, duration 17787ms, Failure. Started 23:36:37.157Z, duration 15545ms, Failure.
- Both failures: `PreviewAutomationTimeoutError: Preview automation open timed out after 660000ms.` with `[cause]: Error: page.goto: Timeout 15000ms exceeded.`

### Related issues

#22594 (same message, but there the call really did block for 11 minutes), #16921.

---
*Triaged and drafted by Claude Opus 5.5 (claude-opus-5-5) via Claude Code, using `t3 triage`.*