HarnessAgent now supports image inputs for all harness adapters
HarnessAgent can now pass images to vision models through every supported coding-agent adapter, so image-based tasks can use the same AI SDK message flow as text.
HarnessAgent accepts images in the standard AI SDK message format, using a content array with **text** and **file** parts. The feature works across **all available adapters**, including Claude Code, Codex, Cursor, and OpenCode.
Builders can include screenshots or other image files in agent requests without changing the message format between harnesses. The image is handled by the adapter’s underlying vision model.
HarnessAgent accepts images in the standard AI SDK message format, using a content array with **text** and **file** parts. The feature works across **all available adapters**, including Claude Code, Codex, Cursor, and OpenCode. Builders can include screenshots or other image files in agent requests without changing the message format between harnesses. The image is handled by the adapter’s underlying vision model. Image understanding still depends on the underlying model. The announcement does not specify supported image formats or size limits.
This makes image input part of HarnessAgent’s common message format across its available adapters, so builders can send screenshots or other images without changing formats when switching harnesses. Image understanding still depends on the underlying vision model, and the source leaves supported formats and size limits unspecified.