dsh-cooldown-retry: Upstream 429 Cooldown Wait and Fault-Tolerant Retries for DeepSeek Harness

dsh-cooldown-retry is a plugin for DeepSeek Harness that addresses the issue where, when a shared gateway returns a 429 error, the built-in mechanism only falls back to local backoff, causing requests to fail. It provides a more patient automatic retry mechanism by reading delay hints such as `retry_after_seconds` from the response body. In addition, the plugin can absorb 400 errors incorrectly labeled as “audio modality not supported,” manage retry budgets, and perform cancellation-aware waiting. After installation, the plugin entry should be configured after the `llm-retry` entry, and it supports customizing parameters such as the maximum retry count and delay window to improve workflow availability in unstable gateway environments.

Read More