Introduction

In using DeepSeek Harness (DSH), configuring the retry policy (retryPolicy) for a model provider usually requires editing the settings.yaml file directly. This approach lacks visual feedback, and adjusting parameters (such as retry mode, retry count, and backoff strategy) can be error-prone. The dsh-retry-policy-settings plugin aims to solve this issue by providing a visual editing interface.

Plugin Overview

This is a DeepSeek Harness (DSH) plugin maintained by guanosine-diphosphate. It allows direct editing of a model provider’s retryPolicy configuration in the DSH Web UI, including mode, retry count, retryable error codes, and exponential backoff parameters, and writes changes to the configuration file immediately.

Core Features

  • Auto Discovery: Automatically reads all registered model providers (such as custom OpenAI-compatible APIs, the official DeepSeek provider, etc.).
  • Full-Field Editing: Edit all optional fields of a provider’s retryPolicy (mode, maxRetries, retryableCodes, backoff parameters, etc.). Fields that have default values in the file are also displayed.
  • One-Click Restore: Supports one-click removal of the configuration and restoration of defaults.
  • Environment Adaptation: Chinese/English automatically follows DSH’s global language switching, with adaptive dark/light theme support.

Installation

Install it from GitHub into your DSH profile:

dsh plugin --profile web add "github:guanosine-diphosphate/dsh-retry-policy-settings#main"

After installation, add dsh-retry-policy-settings to the profile’s dsh.profile.bundles (<profile>/package.json), then restart the Web UI to use it.

Usage

  1. Open the DSH Web UI.
  2. Go to Settings in the lower-left corner.
  3. Click Model Retry Policy (displayed as Model Retry Policy in the English interface).
  4. Select the specific provider, adjust the mode and parameters, and click Save.

Configuration Write Location

The configuration modified by the plugin is written to settings.yaml. The specific path depends on the adapter that the model provider belongs to:

Provider Type settings.yaml Path
Custom / pi-ai provider (e.g., scnet) llm-pi-ai.providers.<id>.retryPolicy
Official DeepSeek (deepseek-official) llm-deepseek.retryPolicy

Dependencies and Notes

  • Dependency: Depends on the official plugin @deepseek-ai/dsh-llm-retry (enabled by default in DSH).
  • Mode Limitation: In always mode, the maxRetries and retryableCodes fields are disabled (due to underlying Schema constraints).

Conclusion

This plugin provides DSH users with a convenient entry point for managing model retry policies, suitable for developers who need to frequently adjust retry logic in the Web UI.

GitHub Repository