pptx
Warn
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [DYNAMIC_EXECUTION]: The helper script
scripts/office/soffice.pydynamically generates a C source file at runtime in the system's temporary directory. It then executesgccto compile this source into a shared library (lo_socket_shim.so). This library is subsequently injected into thesoffice(LibreOffice) process using theLD_PRELOADenvironment variable to intercept and shim system calls. While this is used to bypass socket restrictions in sandboxed environments, the use of runtime compilation and injection is a high-capability pattern. - [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied data through multiple ingestion points, most notably in
scripts/html2pptx.jswhere HTML files are rendered using the Playwright browser automation library. - Ingestion points: User-provided HTML slide content and PPTX files (scripts/html2pptx.js, scripts/inventory.py).
- Boundary markers: Absent; external content is rendered or parsed without strict isolation or "ignore instructions" delimiters.
- Capability inventory: The skill has extensive capabilities including arbitrary shell command execution via
subprocess.run(scripts/thumbnail.py, ooxml/scripts/pack.py), file system write access, and network access via Node.js dependencies. - Sanitization: There is no evidence of HTML sanitization before rendering, which could allow for script execution or data exfiltration within the browser context.
- [COMMAND_EXECUTION]: Multiple scripts use
subprocess.run()to execute system utilities includingsoffice,pdftoppm,git, andgcc. While these calls are tied to the primary functions of converting, thumbnailing, and validating presentations, they represent a significant attack surface if the input parameters (such as filenames or paths) are manipulated by a malicious actor.
Audit Metadata