Preface

DeepSeek Harness (DSH) may sometimes encounter request body structure mismatches when calling custom OpenAI Chat Completions endpoints. Simple connectivity checks may pass, but Harness’s default request structure (such as the developer role and the max_completion_tokens field) may not be accepted by the target endpoint. DSH Provider Passport is an independent community plugin designed to discover and remediate these request dialect differences before the first real task fails in Harness, through a bounded, review-first preflight mechanism.

Plugin Overview

Project name: armywas/dsh-provider-passport
Maintainer: ArmyWas
License: MIT
Project type: Model inference plugin

This is a public preview plugin, independent from DeepSeek official, and intended only to describe compatibility.

Core Capabilities

1. Request dialect preflight
The plugin checks request differences between custom endpoints and Harness. It only proposes compatibility fields supported by Harness and validates them through the actual Harness LLM runtime. If validation fails, the plugin automatically restores the previous settings, and cancellation does not write configuration changes.

2. Security and privacy
* Memory safety: Credentials and custom headers are retained only in memory and never enter reports or logs.
* Request limits: Each preflight sends at most 10 fixed requests, each request allows at most 1 output token, and each request has a 12-second timeout.
* Minimal impact: Only the selected model receives the minimal observed compatible profile; other models are unaffected.
* Explicit confirmation: Every preflight requires explicit user confirmation, and no background tasks are executed.

3. Scope limitations
* Routes must explicitly declare api: openai-completions. When api is omitted, the plugin excludes the route instead of guessing.
* Proposed fields are limited to maxTokensField, supportsDeveloperRole, supportsStore, supportsReasoningEffort, and supportsUsageInStreaming.
* Routes that use Responses, Anthropic Messages, or other protocols are out of scope.

Installation and Enablement

Before use, ensure your environment meets the following requirements:
* Node.js 22.19+
* DeepSeek Harness Web profile

Run the installation command:

npx @deepseek-ai/dsh plugin --profile web add dsh-provider-passport@preview

After installation, start Harness Web:

npx @deepseek-ai/dsh web

Usage Flow

1. Open the plugin interface
Open Settings → Plugins, then expand Provider Compatibility Passport.

2. Select a model and run preflight
Select a custom OpenAI Chat Completions model. Review the request budget (at most 10 requests, 1 token per request, 12-second timeout), and confirm the preflight.

3. Validation and reporting
The plugin validates changes through the runtime. After validation passes, click Copy sanitized report. The report removes endpoints, model IDs, credentials, request text, response bodies, and model outputs, making it suitable for sharing with the community.

Ecosystem Context and Considerations

DSH’s philosophy is “everything is a plugin.” This plugin is only responsible for last-mile compatibility checks. It does not replace general API, streaming, or tool compliance tools, and it does not cover routing, fallback, or proxy adapters.

Before using it, we recommend reading the source code and checking the license. For tests involving sensitive enterprise endpoints, read the privacy design documentation first.

Summary

DSH Provider Passport provides a secure, controlled way to identify protocol differences between custom OpenAI-compatible endpoints and Harness. Through runtime validation and automatic rollback, it reduces the cost of misconfiguration.

For more information, visit:
* Plugin catalog
* GitHub repository