The Agent Skills Directory

[PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection by extracting and processing text from untrusted external images or documents.
Ingestion points: Untrusted data is introduced through the --file-url and --file-path parameters passed to scripts/ocr_caller.py.
Boundary markers: The skill lacks specific delimiters or instructions that would prompt the agent to ignore potentially malicious instructions found within the extracted text.
Capability inventory: The skill relies on python to execute the scripts/ocr_caller.py script and requires the agent to parse the resulting JSON data.
Sanitization: There is no process defined to sanitize or validate the extracted text before it is integrated into the agent's context.

read-ocr