SKILL.md
Bloom
The world model
Bloom is the system that owns brand. When you call bloomgenerateimage or bloomeditimage, the brand's visual DNA (palette, typography, tone, aesthetic, logo treatment) is already attached as image references, and a generation system prompt that knows the brand is applied on top, enforcing quality stances like restraint, whitespace, and no tech clichés.
So your prompt's job is narrow: describe the subject, the composition, and the medium. That's where your prompt adds information; everything else is already handled.
Where to look
Read the rule for the task you're on:
| Task | Rule |
|---|---|
| Writing the prompt text | [rules/prompting.md](rules/prompting.md) |
| Writing the headline copy on the image | [rules/copy.md](rules/copy.md) |
| Setting up a run (references, model, batch) | [rules/workflow.md](rules/workflow.md) |
| Choosing an aspect ratio | [rules/channels.md](rules/channels.md) |
References
- For tool semantics (parameters, async behavior, reference images, workspaces, credits), the tool descriptions on
bloomgenerateimage,bloomeditimage,bloomlistimages, etc. are the source of truth. Read those rather than restating them here. - Bloom has no memory between calls. If your prompt references an image (with a pronoun, a definite article, or "this"), that image has to be attached. The MCP server's
instructionsfield covers the architectural details.