mobile: quota errors show generic "run stopped with an error" instead of the provider error message #64079

Summary

When a session run fails because of a provider quota error, the mobile app shows a generic toast ("the run stopped with an error") instead of the human-readable error the server already delivers (type provider.free-limit, message "Free usage limit reached. Connect a provider or subscribe to OpenCode Go to keep going, or switch models.", including retryAfterMs). The user has no way to troubleshoot from the phone.

Environment

• opencode version: 2.2.6
• OS: macOS (Darwin 27.0.0, arm64) — server; client is the OpenCode mobile app paired via opencode pair --remote
• Terminal: unavailable (server-side only)
• Shell: /bin/zsh
• Install/channel: latest
• Active plugins: usage-band, skillspector, side-chat

Reproduction

1. Pair the mobile app with the desktop server (opencode pair --remote).
2. Use a chat whose model is a free OpenCode model (e.g. opencode/glm-5.3-flash-free) until the free usage limit is hit (HTTP 429).
3. Send a message from the phone app.

Expected Behavior

The app shows the actual cause, e.g. "Free usage limit reached. Connect a provider or subscribe to OpenCode Go to keep going, or switch models." — ideally with the retry time (retryAfterMs ≈ 10.4 h in my case) or a link to the suggested action (switch model / subscribe).

Actual Behavior

The app shows only: "the run stopped with an error". No reason, no retry time, no action.

The structured error is available server-side and to clients; GET /api/session/{id}/message returns an idle message with:

{
  "type": "idle",
  "outcome": "failed",
  "error": {
    "type": "provider.free-limit",
    "message": "Free usage limit reached. Connect a provider or subscribe to OpenCode Go to keep going, or switch models.",
    "status": 429,
    "retryAfterMs": 37603000
  }
}

The assistant message in the same session carries the same error object, and the server log records it ("Failed to drain Session", cause AI.Error.QuotaExceeded). So the data needed for a good error message is already being persisted/streamed; the mobile client just does not surface it.

Additional Context

• Reproduces consistently for any run that ends with this quota error.
• Workaround on the user side: pick a non-free model (or wait out the limit reset); the generic toast still appears for any future failure regardless.
• Screenshot of the toast available on request; logs redacted of tokens/URLs.