gemini-cli - 💡(How to fix) Fix BUG: Client-side 429 routing loop causes runaway token consumption and exhausts billing caps

What happened?

The Setup: Running gemini-cli with a local Home Assistant MCP server on a paid Tier 1 AI Studio key.

The Flaw: When hitting server-side capacity limits, the NumericalClassifierStrategy.route throws a 429 error. Instead of gracefully failing or alerting the user, the client enters an invisible, infinite background exponential-backoff loop.

The Impact: On every single background retry, it repeatedly compiles and bundles the massive Home Assistant schema into the context window, burning through millions of tokens in minutes until it slams into the billing cap, all while the console interface sits entirely frozen on "Thinking...".

What did you expect to happen?

My requests should have been handled correctly, I was within quota the whole time.

Client information

bug-report-history-1779037725185.json

Run gemini to enter the interactive CLI, then run the /about command.

> /about
# paste output here
 CLI Version                              0.42.0

</details>

### Login information

I burned through millions of tokens on Google Account and API.

### Anything else we need to know?

This is a real bad reputational issue when you're billing people for a fault in your own tools.