Preface

Running agents in DeepSeek Harness (DSH) often encounters friction not in “whether the model can be called,” but in how to log in with providers, switch default models, handle images with text-only models, and whether token usage and costs can be clearly viewed in the native interface. In the community, some use API Keys to fill configurations, some manually piece together OAuth callbacks, and others split usage statistics into external scripts—each approach works, but maintenance costs are not low.

dsh-AuthInOne is a community-maintained DSH plugin that consolidates provider login, model management, optional visual fallback, usage and cost statistics, and web restart on the same port into a self-contained installation process. Below, we introduce it following the structure: “What it is → What it can do → How to install → How to use → What to note.”

What Is This

dsh-AuthInOne (GitHub: Stormycry-cryp/dsh-AuthInOne) targets DSH’s Models and Usage settings pages, offering:

  • Provider/account login (including OpenAI Codex, Kimi Code, and several experimental compatibility flows)
  • API Key and custom OpenAI-compatible provider configuration
  • Default model switching for future sessions
  • Optional visual fallback when the primary model declares text-only input
  • Cross-session token usage and auditable cost statistics
  • Post-installation web restart on the same URL and port, without modifying DSH core or manually editing profiles

Maintainer: Stormycry-cryp. Current version: Alpha v0.2.0-alpha.4, compatible with official DSH 47f9438. License: MIT. The project is community-maintained, not an official product of DeepSeek or OpenAI.

SkillHub catalog categorizes it under “Model Inference.” As of writing, it has approximately 64 stars on GitHub.

Core Features

The following capabilities are derived from the project README’s Verified capabilities and account support matrix, listing only those with verified status.

Provider and Account Login

OpenAI Codex (Stable): Browser OAuth using state, S256 PKCE, and loopback callback; verification reaches the auth.openai.com user confirmation boundary.

Kimi Code (Experimental): Full authorization link based on RFC 8628 device flow; the link embeds a short code, and users typically only need to log in and confirm.

Experimental Compatible Accounts (all marked as Experimental compatibility, verified to each provider’s authorization boundary): xAI Grok, Anthropic, GitHub Copilot, Command Code, Cursor, Google Antigravity, Kiro.

Credential handling principles: OAuth verifier, authorization codes, device codes, access tokens, and refresh tokens remain in the Host; the plugin does not import ~/.codex/auth.json, OpenCodex state, cookies, browser storage, OTP, or other product credentials.

API Key and Custom Provider

Supports managing API Key providers (including DeepSeek API Key with live verification) and custom OpenAI-compatible Base URLs, request headers, and model mappings via the DSH native Models page; AuthInOne preserves native Models cards and reads their public provider projections.

Plan/API presets cover OpenAI, xAI, Gemini, Anthropic, Kimi Code, GLM Coding Plan, ModelStudio/Qwen, etc.; usage restrictions for GLM and Qwen are displayed in the interface.

Model Switching and Visual Fallback

Allows setting the default model for future sessions without replacing DSH’s per-session model selector, with support for connection testing.

When the selected primary model explicitly declares text-only input, a visual model can be configured separately to describe images only when needed. Supports PNG, JPEG, WebP, GIF via DSH ImageBlock references; multiple images, native multimodal passthrough, and fallback disabling paths have corresponding test coverage.

Usage and Cost Statistics

Under Settings → Usage, provides cross-session token activity, model/provider-dimensional statistics, time buckets, and cost projections. README notes: Reconstructed KPIs such as 26,383 Tokens, heatmaps, and other projections based on real DSH session logs. Some providers (e.g., Codex, Kimi, xAI, Anthropic, Cursor, Antigravity) support subscription quota display when upstream data is available, on a best-effort basis.

Self-Contained Installation on the Same Port

Installation commands are provided by this package: verify the running standard DSH Host, proceed via the official dsh plugin add path, then replace that Host on the same URL. No patched DSH core, manual profile editing, second web instance, or temporary port is required.

Installation and Enablement

The environment must have pnpm installed and a running standard DSH Web profile on the machine.

Official installation command (README documentation version, corresponding to v0.2.0-alpha.4):

pnpm dlx github:Stormycry-cryp/dsh-AuthInOne#v0.2.0-alpha.4 install --profile web

Note: Execute the above command as one complete operation; do not split it into manual dsh plugin add followed by ad-hoc restarts. The bootstrapper verifies the listener, installs via the official DSH plugin CLI, returns to the original URL, and reports ready before accepting UI operations. If the bootstrapper rejects due to an unknown launcher or DSH owner fingerprint, stop and follow the error message—do not attempt to guess workarounds.

For non-default ports, source code checkout installation, and upgrade/removal methods, see the “Install, upgrade, and remove” section in the GitHub README.

Typical Usage

After installation and confirming web readiness:

  1. Open DSH and navigate to Settings → Models or Settings → Usage.
  2. On the Models page: Complete account authorization for Codex/Kimi as needed, or fill in API Keys/custom OpenAI-compatible providers; set the default model for future sessions; configure visual fallback models if necessary.
  3. On the Usage page: View token activity, usage distribution by model and provider, and cost statistics; when upstream quota data is available, subscription quotas can be viewed on the Models page (best effort).

The browser receives aggregated usage, anonymization status, authorization URLs, and user codes when needed; sensitive tokens do not leave the Host.

Suitable Scenarios and Notes

Suitable for:

  • Those who wish to complete OpenAI Codex or Kimi Code login within DSH, or explicitly accept experimental compatibility flows.
  • Unified management of API Keys and custom OpenAI-compatible providers on the native Models page.
  • Need for default model switching, image fallback for text-only primary models, and reconstructible token/cost statistics.

Notes:

  • The plugin runs with the current DSH process permissions; please review the source code and MIT license yourself before installation.
  • The seven additional account flows are all experimental compatibility, behavior may change with upstream updates; OpenAI Codex is the stable path noted in the README.
  • Minimum compatibility: Most account and visual fallback capabilities require official DSH 47f9438 and bundled compat owner; some API Key capabilities are verified in DSH 0.1.0-rc.6.
  • SkillHub is an independent community directory, with no official affiliation to DeepSeek or High-Flyer; the plugin is also a community project.

Links

If this plugin is useful to you, you can star the project on GitHub to support maintenance.