image-pipeline

name	image-pipeline
description	Owns MCP tool registration, provider adapters, profile-first model selection, storage, and local planning for image generation/editing.

image pipeline

Owns

• MCP tool registration
• provider adapter boundaries
• curated profile-first model selection
• file naming and persistence contract
• local implementation planning for generate_image and edit_image

Invariants

• server.py should stay thin and provider-agnostic
• curated profiles should be the default UX; raw model ids are override-only
• storage.py owns final path creation and atomic-ish writes
• adapter modules own provider HTTP payloads and response parsing
• successful tool results are structured objects with path, mime_type, provider, model, and profile

Current local tasks

1. keep curated profiles aligned with live provider reality
2. keep reference-image semantics deterministic and role-based
3. close Gemini live proof once GEMINI_API_KEY is available

Safe change order

1. update the contract if the agent-facing shape changes
2. update one provider adapter
3. keep storage behavior centralized
4. prove with tests and smoke