The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided files (Excel, Word, PDF, Text) to extract content for analysis. This creates a surface for indirect prompt injection if the document content contains malicious instructions intended to subvert the AI agent's behavior during the expansion phase.\n
Ingestion points: scripts/convert_to_markdown.py reads input files from the local filesystem.\n
Boundary markers: Absent; the agent reads the full output of the conversion script.\n
Capability inventory: Access to Bash for running Python scripts and Write for file creation.\n
Sanitization: Relies on standard document parsing libraries (pandas, PyPDF2, python-docx) for text extraction.\n- [DYNAMIC_EXECUTION]: The SKILL.md instructions direct the agent to generate and run temporary Python scripts to handle data transformation (e.g., converting list data to Excel using pandas). This is a controlled use of dynamic execution based on pre-defined templates for local data manipulation.

process-faq