[FEATURE] Allow `apiKeyHelper` to trigger when it is set and claude gets a 403 error from the LLM provider

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

I am encountering a timing issue with oauth token's getting refreshed as part of my apiKeyHelper. Seems like the internal flow is like this:

1. claude code runs my apiKeyHelper, which returns a token that will expire in 10s
2. claude code makes inference API calls for >10s
3. -> claude code gets a 403 from the inference provider

but because my apiKeyHelper is refreshing an oauth token, if it was simply re-ran then the 403 would resolve

My current workaround is to restart claude code, which is fine for eager users but will feel buggy for regular users in my org.

Proposed Solution

(optionally) run apiKeyHelper when receiving 403 errors from inference api calls

Alternative Solutions

_No response_

Priority

Medium - Would be very helpful

Feature Category

API and model interactions

Use Case Example

_No response_

Additional Context

_No response_