Preface

DeepSeek Harness (DSH) binds conversational Agents and extensible tools together. To generate images directly within a session or perform editing using multiple reference images, one typically needs to integrate HTTP APIs manually, manage credentials, and handle attachment uploads. DSH FHL Image is an independent, community-maintained plugin (MIT) that integrates FHL Images’ generation, editing, Worker pools, and a chat-configurable credential workflow into DSH. It is not an official DeepSeek or FHL product; you must have DSH installed and running before use.

What is this

  • Name & Package: DSH FHL Image, npm package @fhl-plugins/dsh-fhl-image, current plugin version 0.2.2.
  • Maintainer: supart. Source code and distributions are available on GitHub.
  • Category: Model Inference (Community Directory SkillHub Entry).
  • Version Correspondence: 0.2.2 targets the official desktop DSH 0.2.0-rc.2 package line; engines.dsh is set to >=0.2.0-rc.2 <0.3.0-0. DSH validates the peer dependency range upon installation or startup. Mismatches will trigger an incompatible-version error; in this case, please switch to a compatible plugin version rather than forcefully bypassing the check.

Core Features

The plugin exposes three tools to the Agent (the specific names depend on the session’s tool list):

  1. fhl_image_configure: When explicitly requested by the user, configures up to ten Workers using Keys provided in the chat; upon success, it only reports the sanitized count of Workers.
  2. fhl_image_generate: Generates one to nine images via the FHL Images API.
  3. fhl_image_edit: Performs editing or combination tasks using one to ten ordered reference images, capable of producing one to four independent variants.

Other behaviors (detailed in the README): Supports up to ten Workers; credentials are referenced via DSH credential providers as FHL_IMAGE_API_KEY through FHL_IMAGE_API_KEY_10; supports Worker cooldown retries, per-Worker authentication isolation, cancellation, bounded responses, and credential sanitization; generation results return to the conversation via the DSH attachment service.

Installation and Activation

The installation package is the Release attachment fhl-plugins-dsh-fhl-image-0.2.2.tgz (.tgz is a plugin package and cannot be run by double-clicking). You must use a direct download link ending in .tgz; repository URLs or releases/latest will be treated as git sources, but the repository does not contain the build output lib/, nor can @fhl-plugins/dsh-fhl-image be installed via npm.

Direct Download Link:

https://github.com/supart/DSH-FHL-Image-Plugin/releases/download/v0.2.2/fhl-plugins-dsh-fhl-image-0.2.2.tgz

One-Prompt Installation (Recommended): Send the following block in a DSH session to let the Agent install it using plugin management tools:

Please use the plugin management tool to install this FHL image plugin package:
https://github.com/supart/DSH-FHL-Image-Plugin/releases/download/v0.2.2/fhl-plugins-dsh-fhl-image-0.2.2.tgz

After installation, turn on its enable switch, then let me completely quit and reopen the App.
Then start a new session and confirm that fhl_image_configure, fhl_image_generate, and fhl_image_edit are in the tool list.

Installation will modify the desktop profile’s package.json, pnpm-lock.yaml, and node_modules/, requiring danger-full-access or authorization for the current run. The desktop App exclusively uses the desktop profile; do not execute dsh plugin against it in an external shell.

GUI Method: Place the tarball in a directory without spaces. In the sidebar “Plugins” page, install using the absolute path → Turn on the enable switch → Completely quit and restart the App → Start a new session to verify the three tools.

CLI DSH (Advanced, Custom Profile):

dsh plugin --profile fhl-image add /absolute/path/to/fhl-plugins-dsh-fhl-image-0.2.2.tgz
dsh --profile fhl-image --dump-config
dsh --profile fhl-image

If tools do not appear: Verify they are enabled, restart the App or Host, and start a new session. After modifying or reinstalling the plugin, old processes may still expose the old tool catalog; a restart is mandatory.

Typical Usage

Configuring Workers via Chat

After starting a new session and confirming the three tools exist, you can configure the API first (do not call image generation tools during the configuration phase):

I want to configure FHL image generation APIs. Please use the fhl_image_configure tool and first tell me how to provide one or more Worker Keys; do not call the image generation tool yet.

Paste the Keys into the chat box as instructed by the Agent, one per line. The raw Keys will pass through the current model with the user message and may remain in the DSH session history. If this is unacceptable, you should use the DSH credential mechanism to write directly to reference names like FHL_IMAGE_API_KEY, FHL_IMAGE_API_KEY_2, etc.; the config item apiKeyEnv is merely a prefix for these reference names, not an environment variable.

Image Generation

Please use the FHL image generation tool to generate a full-color rendering of a minimalist modern building. Generate 1 image, aspect ratio 16:9, resolution 2K.

Editing

Drag reference images into the chat box, for example:

Please use this reference image for editing, retain the subject's features, and generate a full-color rendering.

When sources is omitted, the plugin attempts the following in order: images in the current message → earlier user uploads → the most recent image generated by this plugin within the same session and process. After restarting the App, the third option becomes invalid; you must explicitly pass the attachmentId from the previous result.

Overriding Plugin Configuration

Defaults are set at the bundle layer and do not read environment variables. You can override them at the profile patch layer, for example:

# ~/.dsh/profiles/desktop/cordis.patch.yml
- id: fhl-image
  name: '@fhl-plugins/dsh-fhl-image'
  config:
    baseURL: https://www.fhl.mom
    timeoutMs: 180000
    workerCooldownMs: 60000

Use Cases and Notes

Suitable for users already on the DSH Desktop 0.2.0-rc.2 line who need FHL Images generation and multi-reference editing, and wish to use a Worker pool to distribute requests. The plugin runs with the current DSH process permissions. You should review the source code and MIT license before installation and assess the risks regarding API Keys and network egress yourself. The community directory SkillHub has no official affiliation with DeepSeek or High-Flyer. The facts in this article are subject to the project README and Releases.

Links

  • Community Directory: https://www.skillhub.cn/plugins/supart/DSH-FHL-Image-Plugin
  • Source Code & Documentation: https://github.com/supart/DSH-FHL-Image-Plugin