The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill extracts text from external, untrusted documents such as PDFs and scans, which may contain malicious instructions designed to influence agent behavior.
Ingestion points: Untrusted data enters via scripts/extract_marker.py, scripts/extract_pymupdf.py, and the web_extract platform tool.
Boundary markers: No explicit delimiters or instructions are provided to the agent to disregard content extracted from these documents.
Capability inventory: The skill possesses the ability to write files (extract_images), execute commands through helper scripts, and perform network operations via platform tools.
Sanitization: Extracted document content is not sanitized or filtered before being processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill instructs the user to install well-known Python packages from official registries and notes that the marker-pdf tool downloads approximately 2.5GB of AI models from Hugging Face during its first run.
[COMMAND_EXECUTION]: The skill relies on the execution of Python scripts and CLI tools (e.g., marker_single, pip) to perform document processing and setup tasks.

ocr-and-documents