The current implementation of rate limit handling results in rate limit updates only on exit from the CEL evaluation. This has the unfortunate consequence that any complex API request chain may deplete the token bucket of the rate limiter and exceed the API's rate limits before being updated.
We have a change merged into the mito repository that enables immediate updating of the rate limits to avoid this. That change needs to be wired into the CEL input to make it available to integrations that need to properly handle rate limits.