[rgarcia]

Summary

Lets the PR Risk check run on federated model credentials instead of stored API keys.

kernel/security-workflows#36 changes the reusable pr-risk.yml so its classify job requests id-token: write. It exchanges the job's GitHub OIDC token for short-lived Anthropic/OpenAI credentials. A reusable workflow job can't request more permissions than its caller grants, so this caller grants id-token: write.

Merge this before kernel/security-workflows#36. It changes nothing with the current pr-risk.yml, which ignores the extra permission. If #36 merges first, PR Risk fails to start here until this lands.

The ANTHROPIC_API_KEY / OPENAI_API_KEY secrets are still passed. #36 ignores them, and they'll be removed in a follow-up once every caller has switched.

Testing

kernel/security-workflows#36 was tested end to end: a caller granting id-token: write called the new pr-risk.yml on a throwaway PR with no API keys passed, and the classifier ran and posted its comment.

[rgarcia]

Picked tnsardesai to review these related PRs (kernel/security-workflows#36, kernel/kernel#5359, #550, kernel/infra#2088, kernel/kernel-images-private#584): touched the changed code in workflows; 4 pending