We use ChatGPT Skills in a Business workspace to standardize professional image-editing workflows for a tattoo studio.
We have multiple specialized Skills, for example:
- extracting a tattoo motif from a photograph,
- creating a tattoo stencil,
- coloring an existing tattoo design,
- placing an existing design on a model.
Each Skill contains task-specific instructions that materially affect how the image-generation tool should be used.
Currently, when a user uploads an image and writes a request such as “Create a stencil from this image”, ChatGPT frequently invokes the image-generation tool directly without first checking or loading the installed Skill whose name and description clearly match the request. We tested it within the browser (chrome, firefox) as well as the app (Windows, iOS).
Explicitly selecting the Skill via @mention works correctly.
This makes automatic Skill invocation unreliable for image-processing workflows, because image-generation routing appears to happen before Skill routing.
Suggested improvement
Allow Skills to declare that they apply to image generation or image editing and must be evaluated before the image-generation tool is invoked.
For example, Skills could support metadata conceptually equivalent to:
capability: image_editing
before_tool: image_gen
or a priority such as preferred / required.
Desired routing:
User request → identify relevant installed Skills → load applicable Skill instructions → select/invoke image-generation tool.
This would allow organizations to enforce standardized image-processing workflows while still using ChatGPT Images as the underlying generation tool.