AI Agent Hub
Back to plugins
⚙️

dsh-retry-plugin

Workflow Updated 2026.08.22

Run the following command in DeepSeek Harness:

dsh plugin install jedzqer/dsh-retry-plugin

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install jedzqer/dsh-retry-plugin in your terminal to install; the source repository is at https://github.com/jedzqer/dsh-retry-plugin . After installation, configure the retry policy in DSH Settings.

About this plugin

Working with DeepSeek Harness (DSH), network glitches, rate limits, 5xx responses, and quota exhaustion can interrupt a conversation at any moment. DSH's built-in retry covers five transient error codes with exponential backoff and persistent events, yet abandons QUOTA, CONTEXT_WINDOW_EXCEEDED, and similar failures outright. The unbounded always-retry mode, meanwhile, can burn through quota repeatedly. dsh-retry-plugin fills that gap: retry any error, whitelist-only, or blacklist-except—three strategies let you decide exactly which failure codes deserve another attempt, with a hard budget of 0 to 50 retries per failure.

The core experience is configurable, bounded, and instantly effective. Tune the initial delay, maximum delay, and jitter ratio of the exponential backoff; provider Retry-After headers are honoured automatically. Every change writes to the plugin's own settings namespace and hot-reloads on the next request—no restart needed. The retry itself rides DSH's native same-turn resubmission path, complete with persistent llm/retry events, cancellable backoff waits, and a live countdown in the chat stream, without injecting stray continue echo messages into conversation history.

If you hit rate limits and transient faults in high-concurrency or long-session workflows, want to exclude auth-class errors to protect quota, or prefer to include quota-class errors for a recovery chance, this plugin gives you precise control layered on top of DSH's native capabilities. Pair it with dsh-sound-plugin and you will hear a notification the moment the AI finishes retrying or needs your attention, even after you have stepped away from the window.

Use Cases

  • Long sessions where AI frequently hits rate limits or 5xx errors and needs precise control over retryable codes
  • High-concurrency workloads that want to exclude auth errors to protect quota while auto-recovering from transient network faults
  • Customizing backoff parameters (initial delay, max delay, jitter) to match different provider rate-limiting policies
  • Hot-adjusting retry mode and budget without service restarts, effective on the very next request

Best For

  • Heavy DSH users who frequently encounter rate limits and transient faults
  • Developers needing fine-grained control over API quota and retry budgets
  • Ops or config users who want to hot-adjust retry policies without service restarts