pptx
Warn
Audited by Gen Agent Trust Hub on Aug 10, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill frequently invokes system binaries to process Office documents.
- In
scripts/office/soffice.py, the skill writes a C source file to a temporary location and executesgccto compile it into a shared object. - It executes
soffice(LibreOffice) with theLD_PRELOADenvironment variable set to the compiled shim to bypass potential AF_UNIX socket restrictions. scripts/thumbnail.pyexecutespdftoppmfor image rendering.scripts/office/validators/redlining.pyexecutesgit diffto validate tracked changes.- [EXTERNAL_DOWNLOADS]: The skill requires several external dependencies to be installed.
- Instructions in
SKILL.mdspecify installing Python packagesmarkitdown[pptx]andPillow. - It also requires the global installation of several NPM packages including
pptxgenjs,react-icons,react,react-dom, andsharp. - It relies on the presence of system tools such as LibreOffice, Poppler, and GCC.
- [PROMPT_INJECTION]: The skill processes external
.pptxfiles which creates a surface for indirect prompt injection. - Ingestion point:
SKILL.mdandediting.mddescribe usingmarkitdownto extract text from user-provided PowerPoint files. - Boundary markers: The extracted content is not explicitly delimited in the provided instructions to prevent the agent from obeying embedded commands.
- Capabilities: The skill has significant local execution privileges, including the ability to run shell commands and compile C code.
- Sanitization: No sanitization of the extracted document text is performed before it is processed by the agent.
Audit Metadata