foolfu86/dsh-agnes-image

A DeepSeek Harness plugin that adds one host tool, agnes_image, for image generation and image editing with Agnes Image 2.5 Flash.

It is deliberately dependency-free: it imports only Node builtins and uses the injected Cordis context (ctx.tools, ctx.credentials, ctx.attachments, ctx.workingDirectory). There is no build step and no @deepseek-ai/* peer dependency, so a git install works without any allowBuilds approval.

Install

dsh plugin --profile demo add github:foolfu86/dsh-agnes-image
dsh --profile demo --dump-config     # expect a "# == dsh-agnes-image" layer
dsh --profile demo

# remove
dsh plugin --profile demo remove dsh-agnes-image

Credentials

The plugin needs an Agnes API key. It looks up the credential reference AGNES_API_KEY at runtime (first process.env, then the profile's credential store). Configure your own key, for example:

export AGNES_API_KEY=your_key_here

or add it to ~/.dsh/.credentials.yaml. No key is stored in this repository.

What the tool does

agnes_image calls the Agnes images endpoint, saves the result into the Session working directory (default agnes-images/), registers it as an attachment so the image shows inline in the conversation, and returns the durable file path.

Defaults (all overridable per call):

Option	Default
model	agnes-image-2.5-flash (also 2.1-flash, 2.0-flash)
size	1K (2K, 3K, 4K)
ratio	1:1 (plus the common 3:4 / 4:3 / 16:9 / 9:16 / 2:3 / 3:2 / 21:9)
baseURL	https://apihub.agnes-ai.com/v1
apiKeyEnv	AGNES_API_KEY
outputDir	agnes-images

Local reference images (png / jpg / jpeg / webp / gif) can be passed for editing and multi-image composition.

Layout

package.json        # declares dsh.bundle.patch
cordis.patch.yml    # inserts this bundle into the profile layer stack
index.js            # the agnes_image tool

License

MIT