Plugin sends auto model requests when it shouldn't

[Infiland]

Plugin sends too much requests to the auto model (gemini 3 flash), causing slowdown and sending more requests than it should

[cristianmiranda]

Finding: gemini-3-flash requests are Cursor CLI internal behavior

After applying the fix from #9 (never passing --model auto to the CLI), gemini-3-flash requests still appear in the Cursor billing dashboard.

This confirms the requests are not caused by the bridge or the plugin — they are internal sub-requests made by the Cursor CLI itself (likely for routing, preprocessing, or context analysis) before dispatching the prompt to the user's selected model.

The pattern in the dashboard is consistent:

gemini-3-flash            ← Cursor internal preprocessing
gemini-3-flash            ← Cursor internal preprocessing
claude-4.6-opus-thinking  ← actual user request

This is Cursor's own behavior and not something the bridge can control or prevent. The fix in #9 still has value (it prevents the bridge from explicitly requesting auto, which would add even more gemini-3-flash calls on top of Cursor's internal ones), but the root cause of the billing entries is upstream in the Cursor CLI.

Closing the associated PR since the core issue is outside the scope of this project.

[Infiland]

Wow, that's really bad for Cursor, I genuinely thought I was sending additional requests somehow and it didn't make much sense to me. Thanks for clarifying this.

Perhaps if there is a bug tracker for Cursor we can report this, it only makes sense since Cursor provides a public API so why would it waste additional gemini flash requests.