The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from external, untrusted documents provided by the user (PDF, DOCX, and TXT formats). Maliciously crafted documents could contain hidden instructions (such as white-on-white text or metadata) designed to override the agent's behavior or manipulate the legal analysis.
Ingestion points: parse_contract in examples/legal_document_parser.py reads data from file paths provided to the agent.
Boundary markers: No specific boundary markers or 'ignore' instructions are used when passing extracted text to the analysis components.
Capability inventory: The skill performs document parsing, regex-based keyword extraction, and report generation. It does not possess network access or arbitrary command execution capabilities.
Sanitization: No sanitization or filtering of the extracted text is performed before processing.
[EXTERNAL_DOWNLOADS]: The skill documentation in SKILL.md identifies dependencies on third-party document processing libraries including pypdf, python-docx, and textract. These are standard libraries for the stated purpose of legal document analysis.

legal-document-analyzer