pptx
Warn
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [DYNAMIC_EXECUTION]: The script
scripts/office/soffice.pycontains logic to dynamically write, compile, and load a C shared library at runtime. It usesgccto create a binary (lo_socket_shim.so) in the temporary directory and injects it into the LibreOffice process via theLD_PRELOADenvironment variable to shim standard socket system calls. - [COMMAND_EXECUTION]: The skill makes extensive use of the
subprocessmodule to execute system-level commands, includingsofficefor PDF conversion,pdftoppmfor image generation,git difffor change tracking, andgccfor runtime compilation. - [DATA_EXFILTRATION]: Documentation in
pptxgenjs.mddescribes the use of theaddImagefunction with arbitrary remote URLs (e.g.,https://example.com/image.jpg). This allows the agent to initiate outbound network requests to non-whitelisted domains during presentation creation. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided
.pptxfiles. Extracted text from these files (viamarkitdown) is ingested into the agent's context without explicit boundary markers or sanitization, creating a surface for indirect instructions to influence agent behavior. - [EXTERNAL_DOWNLOADS]: The skill requires and documents the installation of several external dependencies from public registries, including
pptxgenjs,react-icons, andsharpvia NPM, andmarkitdownandPillowvia PyPI.
Audit Metadata