dsh-llm-error-retry
Run the following command in DeepSeek Harness:
dsh plugin install liaoyuqing/dsh-llm-error-retry
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install liaoyuqing/dsh-llm-error-retry in the DeepSeek Harness terminal (source: https://github.com/liaoyuqing/dsh-llm-error-retry), then restart the target profile to access the LLM error retry settings section in the GUI.
About this plugin
Every DSH user knows the drill: a 429 rate-limit or 503 overload in the middle of an agent run. The built-in fast exponential backoff handles short-lived blips, but when it comes to quota exhaustion or adapters like DashScope that flatten the status code into the message text, you need precise matching and fine-grained control over sleep duration and retry count. dsh-llm-error-retry fills exactly that gap.
The plugin hooks into the agent/request-error waterfall and matches failures by three forms: raw HTTP status codes, provider-agnostic machine codes (RATE_LIMIT, QUOTA, SERVER, etc.), or response field-equal-value pairs. Each rule carries its own sleep window (100 ms to 1 h) and max-retry budget; once exhausted the error is reported as-is, never swallowed silently. Retries render as native cards in the chat pane in real time, and rules are edited live in the GUI settings page -- saved instantly, persisted across restarts, no profile reboot required.
If you build LLM workflows on DSH, hit rate limits regularly, or need vendor-specific retry policies without hand-rolling waterfall listeners and session-event plumbing, this is the drop-in answer. Zero runtime dependencies, no build step -- install and go.
Use Cases
- Auto-retry LLM requests after 429 rate-limit or 503 overload with a custom sleep duration
- Precisely match DashScope or Bailian adapters that flatten status codes into message text via field-equal-value rules
- Tune sleep windows and retry budgets per error code in the GUI settings page, effective immediately and persisted across restarts
Best For
- DSH users who regularly hit rate limits and need differentiated retry policies
- Engineers building multi-provider LLM workflows with per-vendor error-code retry granularity
- Teams that want a drop-in retry solution without hand-rolling waterfall listeners and session-event plumbing
Related Plugins
A method pack that makes AI coding agents plan against your real baseline, prove completion with fresh evidence, and reduce reworks and unsafe changes.
Turns the DeepSeek Harness session into a captain that builds a durable sub-agent team, splits goals into dependency-aware tasks, and coordinates work via direct messages and a live Web UI.
Gives coding agents design judgment, letting Claude Code, Cursor, and 70+ agents generate and iterate high-quality UI, presentations, and graphics right from the terminal.
Run the Pi ecosystem's plugins on DeepSeek Harness, unmodified, via a compatibility layer that implements Pi's public extension ABI on DSH's native services.