31e2e84b08
AI SDK 7 has no 'media' tool output part; images are { type: 'file', mediaType, data: { type: 'data', data } }. After diff_visual, export_image, or any other image result, the next step's prompt failed validation with AI_InvalidPromptError and the chat run stopped. toModelOutput is now typed against the SDK's ToolResultPart output, and a test validates the resulting tool message with the SDK's own schema.
|
||
|---|---|---|
| .. | ||
| assets | ||
| src | ||
| tests | ||
| AGENTS.md | ||
| package.json | ||
| README.md | ||
| tsconfig.json | ||
| tsconfig.test.json | ||
| tsdown.config.ts | ||
@open-pencil/core
Editor engine for OpenPencil: the CanvasKit (Skia WASM) renderer, Yoga layout, the framework-agnostic editor with undo and selection, the Figma Plugin API compatibility layer, the AI/MCP tool definitions, clipboard and vector conversion, and format-neutral document I/O.
It depends on @open-pencil/scene-graph, @open-pencil/pen, @open-pencil/kiwi, and @open-pencil/fig, and keeps browser DOM out so it runs in the app, the CLI, the MCP server, and headless scripts alike.
Import the root barrel or the targeted subpaths listed in package.json exports, for example @open-pencil/core/io, @open-pencil/core/geometry, and @open-pencil/core/canvaskit.
- Programmable overview and SDK docs: https://openpencil.dev/programmable/
- Source and issues: https://github.com/open-pencil/open-pencil
MIT License.