Extend model error retry logic to all providers

[tsubasakong]

Summary

• Remove the OpenAI-only guard from the retry logic so all model providers benefit from automatic retries on transient server errors (5xx).

Why

Google Gemini (and other non-OpenAI providers) returned 503 "high demand" errors that caused immediate task failure. The existing retry logic (max 2 retries with exponential backoff) was gated behind an isOpenAiLikeRuntimeModel check, excluding all other providers.

Changes

• src/agent/runtime/attempt.ts: Removed the isOpenAiLikeRuntimeModel(modelInfo) condition from isRetryableUpstreamModelError, so hasRetryableServerErrorSignature alone determines retryability.

Testing

• All 649 existing tests pass.
• Build succeeds with no type errors.
• The retry behavior (max 2 retries, 1s/2s exponential backoff) is unchanged — only the provider restriction is lifted.

Checklist

• I ran relevant tests, or the Testing section explains why I did not.
• I updated docs, or confirmed no doc changes were needed.
• I confirmed the PR does not include secrets, credentials, or private data.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
openpocket	Ready	Preview, Comment	Mar 17, 2026 5:33pm