dsh-model-retry

In DSH rc.8, model-retry policy was pushed down into individual provider profiles, so tuning how many times a request is retried meant digging through each provider config with no single entry point. dsh-model-retry adds one non-negative integer row to Settings → General and, with a single revision-fenced settings.mutate call, writes retryPolicy.maxRetries across all active configurable providers at once. Existing backoff and retryable-code fields are preserved, and always policies remain unlimited.

The plugin ships as a standalone external Cordis package and does not patch DSH core. Actual retry execution, backoff timing, and retryable-status-code logic are still owned by the DSH engine; the plugin only manages the shared retry-budget setting. It accepts finite non-negative integers including 0, and discovers the current normal/default policies through llm.providers before writing, so locked or removed entries are left untouched.

Best for developers working across multiple inference providers who want a unified retry budget without hand-editing each profile, or who prefer to adjust maxRetries in bulk during local mock testing and CI pipelines.