doc
Warn
Audited by Gen Agent Trust Hub on Mar 6, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTIONREMOTE_CODE_EXECUTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The
SKILL.mdfile explicitly instructs the agent to usesudo apt-get installfor installing system dependencies like LibreOffice and Poppler-utils. This pattern grants the agent administrative access to the host system. - [COMMAND_EXECUTION]: The skill relies on executing external system binaries.
SKILL.mdprovides shell commands forsofficeandpdftoppm, whilescripts/render_docx.pyuses thesubprocessmodule to run these tools programmatically. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted
.docxfiles, which could be used to exploit vulnerabilities in the document parsers or the rendering engine. - Ingestion points: The
input_pathargument inscripts/render_docx.pyand the workflow described inSKILL.mddefine where external document data enters the environment. - Boundary markers: The skill does not implement any delimiters or "ignore instructions" guards when reading or processing the content of the documents.
- Capability inventory: The skill possesses the capability to execute shell commands (via
subprocess.run), install system packages (viaapt-get), and perform file system operations (writing tooutput/doc/). - Sanitization: While the Python script uses
abspathfor path normalization, it does not perform any content-level sanitization or validation of the OOXML structure before processing it withElementTreeor LibreOffice. - [EXTERNAL_DOWNLOADS]: The documentation guides the agent to download and install several Python packages (
python-docx,pdf2image) and system tools from external repositories during setup.
Audit Metadata