Fixes #13572

Problem

executeWithRetry in packages/model-runtime/src/providers/githubCopilot/index.ts reads the Retry-After header correctly, but caps the delay at 10 seconds via Math.min(retryAfter, 10_000). When GitHub Copilot returns a 429 for quota exhaustion with a Retry-After of several hours, the client silently caps it to 10s and retries up to MAX_RATE_LIMIT_RETRIES times — consuming all retry budget without surfacing the real error to the user.

Solution

Added a QUOTA_EXHAUSTION_THRESHOLD_MS constant (5 minutes). Before waiting to retry, we check if the parsed retry-after exceeds this threshold. If it does, we throw immediately (via mapError) instead of waiting and retrying, so the user sees the quota exhaustion error right away.

const QUOTA_EXHAUSTION_THRESHOLD_MS = 5 * 60 * 1000; // 5 minutes

// If retry-after exceeds the quota exhaustion threshold, surface immediately
if (retryAfter > QUOTA_EXHAUSTION_THRESHOLD_MS) {
  throw this.mapError(error);
}

Testing

• Static analysis: verified the changed path handles quota-exhaustion 429 (retry-after > 5 min) by throwing immediately
• Transient rate limits (retry-after ≤ 5 min) still go through the existing exponential backoff path unchanged