Preprocessing

freeact.preproc.preprocess_prompt

preprocess_prompt(
    text: str, max_image_size: int = 1024
) -> str | Sequence[UserContent]

Main preprocessing entry point. Transforms prompt text into agent-ready content.

Currently delegates to preprocess_attachment_tags to resolve image attachments. Other tag types (e.g. <skill>) pass through unchanged for the agent to handle.

Parameters:

Name	Type	Description	Default
text	str	Prompt text, potentially containing <attachment> and <skill> tags.	required
max_image_size	int	Maximum dimension for images (downscaled if larger).	1024

Returns:

Type	Description
str | Sequence[UserContent]	The original text if no images are found, or a multimodal content list with resolved image data.

freeact.preproc.preprocess_attachment_tags

preprocess_attachment_tags(
    text: str, max_image_size: int = 1024
) -> str | Sequence[UserContent]

Resolve <attachment path="..."/> tags to multimodal content.

Scans text for attachment tags, collects image files from the referenced paths (a path may point to a single file or a directory), and loads each image as binary content. The original text is preserved as the last element of the returned list.

When no attachment tags are found, or none of the referenced paths contain images, the original text is returned unchanged as a plain string.

Parameters:

Name	Type	Description	Default
text	str	Prompt text potentially containing <attachment path="..."/> tags.	required
max_image_size	int	Maximum dimension in pixels. Images exceeding this are downscaled while preserving aspect ratio.	1024

Returns:

Type	Description
str | Sequence[UserContent]	The original text when no images are found, or a list of [label, image, ..., label, image, text] entries where each label is a string like Attachment path="...": and each image is a BinaryContent object.

Note

Directory paths include all images in that directory (non-recursive).