dsh-imagegen

dsh-imagegen

Image-generation plugin for dsh: one imagegen tool that renders images through any OpenAI-compatible image backend, so agents can generate images without per-backend glue. The same tool lists the configured routes, echoes the declared model lists, and live-probes the backends.

Tool verbs

Key generate behaviors:

• The prompt argument is required, and it is the only required argument.
• When the call omits model, the tool uses the route's single declared model. When a route declares no models, the tool resolves the model by a live /models probe. When the declared or probed model list contains several models, the call fails and lists the candidates.
• The tool sends size and negativePrompt verbatim, and only when the call sets them. The seed argument must be an integer, and the tool sends it only when the call sets it. The n argument is an integer >= 1 with a default of 1, and a value greater than 1 requests a batch.
• When the call sets out, the output path is that argument (a file path for n = 1, or a directory marked by a trailing separator or by existing on disk). When the call omits out, the tool writes the files to <process.cwd()>/images/<unixts>_<i>.<ext>. The file extension comes from the image magic bytes (png / jpg / webp).
• The tool checks the written bytes for PNG, JPEG, or WebP magic. When the payload is not a recognized image, the call fails with code: output, and the tool keeps the file on disk. When the path cannot be written, the call fails the same way.
• The generate timeout is 180 s by default, and a timeout argument of 1-3600 s overrides it per call. The probe timeout is 10 s.

The plugin ships a dsh-imagegen skill. The skill is the full tool reference for agent contexts: it carries the parameter table, the env contract, and the error codes.

Installation

dsh-imagegen is an out-of-tree plugin for a dsh profile. Install it with the dsh plugin command:

dsh plugin --profile <name> add dsh-imagegen

Configuration is layered: built-in defaults, the IMAGEGEN_* env contract, a cordis patch file, a dsh-imagegen settings section (see below); the plugin creates no config or state files.

Configuration

Providers are configured in any of four layers, lowest to highest priority; the providers verb reports the originating layer of every field, so a shadowed value is never silent:

1. the built-in default (api: openai-compatible),
2. the IMAGEGEN_* env contract (static per process boot),
3. the cordis config layer (a patch entry with id dsh-imagegen and a config object, or profile-level patches),
4. the dsh-imagegen: section in the user settings document ($DSH_HOME/settings.yaml), which the plugin re-reads on every call, so settings edits apply without a restart.

dsh-imagegen:
  defaultProvider: local
  providers:
    local:
      baseURL: http://<host>:<port>/v1
    hosted:
      baseURL: https://api.example.com/v1
      apiKeyEnv: IMAGEGEN_HOSTED_KEY
      models: [flux-1]
      displayName: Hosted Images

The env contract mirrors the config fields 1:1: IMAGEGEN_DEFAULT_PROVIDER, and per route <NAME> the variables IMAGEGEN_PROVIDER_<NAME>_BASEURL, IMAGEGEN_PROVIDER_<NAME>_API, IMAGEGEN_PROVIDER_<NAME>_APIKEYENV, IMAGEGEN_PROVIDER_<NAME>_HEADERS, IMAGEGEN_PROVIDER_<NAME>_MODELS, IMAGEGEN_PROVIDER_<NAME>_DISPLAYNAME. The <NAME> segment is the route name uppercased with - written as _. The headers value is a JSON object string, and the models value is a comma-separated list of model ids. The baseURL value is the OpenAI-compatible base URL including the /v1 prefix (the same shape as an LLM provider base URL); the plugin appends /images/generations and /models to it verbatim, and the image contract has no server-root endpoints. The apiKeyEnv value is a reference to an environment variable name, never a secret in config; per request the value resolves from the process environment, then from $DSH_HOME/.credentials.yaml.

Examples

// Generate with the default route and its single declared model.
{ "verb": "generate", "prompt": "a lighthouse in fog", "size": "1024x1024" }
// Pick a route and model explicitly, and keep the output name deterministic.
{ "verb": "generate", "provider": "hosted", "model": "flux-1",
  "prompt": "an isometric pixel-art garden", "seed": 42, "out": "/tmp/garden.png" }
// Generate a batch into an existing directory.
{ "verb": "generate", "prompt": "three weather variations", "n": 3, "out": "renders/" }
// List what is configured, and where each field comes from.
{ "verb": "providers" }
// List the declared model lists per route (no network).
{ "verb": "models" }
// Probe the backend health (the live /models probe).
{ "verb": "status" }

Troubleshooting

A failing call returns the document { ok: false, error: { code, message, provider?, elapsed_ms } }. The code set is invalid_param / config / network / timeout / http / protocol / output.

• When a config error carries the message "no provider route is configured", no route survived configuration. Check the base URLs; the tool reports each route that configuration drops as a warning in the log.
• When a config error carries the message "credential missing: environment variable "X" is not set", the credential reference does not resolve. Set X in the environment or add it to $DSH_HOME/.credentials.yaml.
• When a config error carries the message "unknown provider "x"", the route name matches no configured route. Route names match case-insensitively and treat _ and - as the same separator; the error lists the configured names.
• When a config error carries the message "multiple providers configured (...) and no default is set", pass the provider argument explicitly, or set IMAGEGEN_DEFAULT_PROVIDER, or set the settings defaultProvider.
• When an output error carries the message "write failed", the tool could not write to the out path. Check the directory and its permissions.
• When an output error carries the message "not a recognized image", the backend returned bytes with no PNG, JPEG, or WebP magic. Inspect the kept file; the backend may return an unsupported container.

Build

pnpm install
pnpm run build     # tsc -> lib/
pnpm run lint      # eslint src/ test/
pnpm run selftest  # in-process: HTTP stub + config matrix + magic-byte checks