docs: distinguish bitmap mockups from vector layout

- Document desktop IMAGE generate/edit versus MCP vector tools
- Note that generate_image is app-only, not miaogai-mcp ALL_TOOLS

Co-authored-by: Cursor <cursoragent@cursor.com>

skills/miaogai/SKILL.md

@@ -239,6 +239,12 @@ Prefer the live tool list from the connected MCP server. Core groups:
 
 > Tool availability can depend on server mode. `open_file`, `new_document`, and disk-writing export paths require `OPENPENCIL_MCP_ROOT` for path scoping. `eval` is available over stdio and disabled in HTTP mode.
 
+## Bitmap mockups vs vector layout
+
+- **Vector layout** — use `render`, `create_shape`, `batch_update`, and related MCP/CLI tools. The result is editable frames, text, and auto layout.
+- **Bitmap mockup** — the desktop editor can generate or inpaint an IMAGE fill (Workflow tab, Image model role, `gpt-image-2`). Chat-only tools `generate_image` / `edit_image` exist in the app; they are **not** MCP `ALL_TOOLS` and not available to `miaogai-mcp`.
+- Do not promise to auto-slice a flat PNG into an icon pack or component tree. Crop slices in the desktop Workflow tab with **Export region** (no extra model call), or `export_image` a selected layer. Rebuild as layout with `render` beside the mockup when the user wants vectors.
+
 ## Key tools for agents
 
 - **`list_documents`** — pick a tab before mutating when several documents are open.