The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting and processing untrusted external documents to influence its analysis and output logic.
Ingestion points: Processes user-uploaded files including .docx and .pdf NDAs, email text, negotiation notes, and .json configuration files.
Boundary markers: The instructions do not define specific delimiters or instructions to ignore embedded commands within the ingested legal text.
Capability inventory: The agent performs file-read and file-write operations across several formats (.md, .docx, .json).
Sanitization: No explicit filtering or sanitization of external content is specified before processing or document generation.
[COMMAND_EXECUTION]: The skill provides detailed technical instructions for the agent to execute Python scripts (utilizing lxml and zipfile) to perform direct XML manipulation within the .docx file structure. This capability is functionally necessary for the skill's specific goal of generating Word-compliant tracked changes but represents a significant execution capability.

nda-abgleich