The Agent Skills Directory

[DATA_EXFILTRATION]: The skill includes logic in references/doc-processing.md to scan the user's ~/Downloads folder using os.path.expanduser. Accessing broad user directories like the downloads folder represents a sensitive filesystem access pattern that could lead to unintended data exposure.
[COMMAND_EXECUTION]: The skill provides Python code snippets (located in references/doc-processing.md) for the agent to execute. These scripts utilize the PyMuPDF (fitz) and pandas libraries to programmatically open and extract text from local PDF and Excel documents.
[PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted external data by processing user-uploaded financial reports (PDF and Excel). This creates an attack surface for indirect prompt injection where malicious instructions could be embedded within the documents. \n- Ingestion points: User-uploaded PDF/Excel documents processed via extract_fund_report and process_fund_holdings in references/doc-processing.md. \n- Boundary markers: None provided in the extraction logic to separate document content from agent instructions. \n- Capability inventory: Filesystem access, file reading, and data extraction using Python libraries as described in references/doc-processing.md. \n- Sanitization: No validation or sanitization of the extracted document content is defined before the information is used by the agent.

invest-fund