impress

Pass

Audited by Gen Agent Trust Hub on Oct 26, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides numerous examples of shell command execution using the soffice (LibreOffice) binary for format conversion, presentation creation, and socket management.
  • [DYNAMIC_EXECUTION]: Python workflows utilize subprocess.run to execute system binaries (unzip, zip) to manipulate the internal XML structure of presentation templates at runtime.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection when processing untrusted presentation files.
  • Ingestion points: The skill reads external presentation files (ODP, PPTX, PDF) and templates (template_path) from the file system.
  • Boundary markers: None identified; external data is interpolated into XML content without explicit delimiters or warnings to the agent.
  • Capability inventory: Includes file system writing (doc.storeToURL, content_file.write_text), subprocess execution for system utilities, and opening local network sockets for UNO communication.
  • Sanitization: The template generation script uses simple string replacement (replace) which does not sanitize for XML structural injection or malicious instructions embedded in the source content.
  • [EXTERNAL_DOWNLOADS]: The skill recommends installing third-party Python dependencies ezodf and odfpy from public registries.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 26, 2026, 08:35 AM