Introduction

When managing multiple provider models in DeepSeek Harness (DSH), it can sometimes be difficult to confirm the connectivity, latency, and input capabilities of a specific model. Existing logs may not be direct enough, or specific diagnostic information may be hard to extract. The dsh-provider-probe plugin aims to solve this problem by providing a manual, low-cost mechanism to test provider routing and adapter configuration.

Plugin Introduction

This is a community plugin named dsh-provider-probe, maintained by Wanbinyu and licensed under the MIT License. It is not affiliated with or endorsed by DeepSeek official.

The plugin sends a minimal real model request through a DSH-registered adapter and only displays the first-token latency, total duration, finish reason, token usage, or a sanitized error. It does not create sessions, write to session logs, or save or display model replies.

Core Features

  1. Routing and Catalog View: Automatically lists the currently active provider routes and adapter model catalog.
  2. Input Capability Display: Shows the text, image, or unknown input capabilities declared by the active model route.
  3. Manual Probe: Supports manually entering model IDs that do not appear in the catalog.
  4. On-Demand Triggering: Sends requests only after the user clicks “Start Probe”; no background polling is performed.
  5. Parameter Limits: Defaults to a limit of 8 output tokens and a forced 20-second timeout.
  6. Result Statistics: Displays first-token latency, total duration, finish reason, and actual token usage.
  7. Error Handling: Preserves the error code, HTTP status, and Request ID, and attempts to remove keys, tokens, and the user directory.
  8. Error Classification and Suggestions: Distinguishes credential, permission, model/address, rate-limit/quota, timeout, network, stream-compatibility, and provider 5xx issues based on explicit status codes and error signatures, and displays next-step suggestions.
  9. Report Export: One-click copies a sanitized diagnostic report suitable for pasting into an Issue/Discussion.
  10. Concurrency Control: Allows only one probe at a time, can be canceled at any time, and avoids duplicate billing and rate limiting.

Installation and Configuration

Before installing, make sure the environment meets the requirements: Node.js >= 22.19 and DeepSeek Harness 0.1.0-rc.6 or later.

Install the plugin:

dsh plugin --profile web add https://github.com/Wanbinyu/dsh-provider-probe/releases/download/v0.3.6/dsh-provider-probe-0.3.6.tgz

Restart DSH Web after installation or update:

dsh web

The plugin supports adjusting parameters through YAML configuration:

- insert:
    - id: provider-probe
      name: dsh-provider-probe
      config:
        timeoutMs: 20000
        maxTokens: 8
        maxMessageLength: 1200

Configuration items:
* timeoutMs: Forced timeout for a single probe, default 20000 milliseconds, range 1000-120000.
* maxTokens: Maximum output tokens for a single probe, default 8, range 1-32.
* maxMessageLength: Maximum number of characters for sanitized errors, default 1200, range 256-8000.

Usage

  1. Open “Settings -> Provider Probe” in DSH Web.
  2. Select a provider and model from the list, or manually enter a model ID.
  3. Click “Start Probe”.
  4. View the result panel to obtain first-token latency, total duration, error classification, and the diagnostic report.

Cost and Privacy

Probe requests go through the normal DSH adapters and the provider configuration already saved by the user, so they may incur a small amount of model cost. The plugin does not read or return API keys; credentials are still resolved by the adapters through their normal path. Model output is discarded, and only latency, usage, and termination status are retained.

Input capabilities come from the declarations of the current DSH model route. No additional image is sent for validation, so there are no extra attachments, network requests, or costs. The copied diagnostic report only uses the route, capabilities, latency, usage, and sanitized error already displayed on the page; it should still be checked manually before publishing.

Error sanitization is a defense-in-depth measure and cannot guarantee coverage of every provider’s custom sensitive formats. Error information should still be checked manually before it is made public.

Applicable Scenarios

This plugin is suitable for the following scenarios: troubleshooting provider connectivity issues, verifying whether a specific model ID is available, testing network latency, analyzing error codes, and preparing diagnostic reports for technical discussions.

Summary

dsh-provider-probe provides a lightweight, manually triggered way to verify DeepSeek Harness provider configuration and model status. With clear error classification and sanitized reporting capabilities, it effectively helps developers locate issues.