oma-image

What problem does it solve? Choosing and authenticating the right AI image generation vendor for each request is tedious, and vague prompts produce poor results. This Skill routes image generation requests to authenticated vendors (Codex gpt-image-2, Antigravity nano-banana, or Pollinations flux/zimage) with prompt clarification, cost guardrails, and reproducible output manifests. ## Core Features & Use Cases - Authentication-aware multi-vendor dispatch: Detects which vendor CLIs are authenticated and routes generation accordingly, with a strict --vendor all comparison mode that writes side-by-side outputs. - Prompt clarification and amplification: Validates that prompts contain subject, setting, style, and aspect-ratio signal, and amplifies brief prompts with user approval before generation. - Reference image support: Forwards up to 10 attached reference images to Codex or Antigravity for style and subject guidance, with magic-byte validation. - Cost guardrails and manifests: Confirms runs estimated at $0.20 or more and writes a manifest.json with prompt, vendor, model, and strategy attempts for reproducibility. - Use Case: A designer asks for a hero illustration in a specific style; the Skill amplifies the prompt, checks vendor health with oma image doctor, generates via the authenticated vendor, and returns the image path plus manifest. ## Quick Start Ask the agent to generate an image, for example: generate a minimalist dashboard hero illustration at 1536x1024 using the default vendor.