The Agent Skills Directory

[PROMPT_INJECTION] (LOW): The skill is susceptible to indirect prompt injection because it is designed to ingest and process untrusted document content.
Ingestion points: ooxml/scripts/unpack.py and scripts/rearrange.py read and extract data from external Office files.
Boundary markers: Absent. The scripts operate directly on document structures without delimiting content for an AI agent.
Capability inventory: ooxml/scripts/pack.py invokes soffice (LibreOffice) via subprocess.run to perform document conversions.
Sanitization: defusedxml is correctly implemented in unpack.py and pack.py to prevent XML External Entity (XXE) vulnerabilities during parsing.
[COMMAND_EXECUTION] (SAFE): ooxml/scripts/pack.py uses subprocess.run with a list of arguments to execute LibreOffice, which avoids shell injection vulnerabilities.
[DATA_EXFILTRATION] (SAFE): No network operations or hardcoded credentials were detected; the skill performs all processing on the local file system.

pptx