deai-office

Fail

Audited by Gen Agent Trust Hub on Jun 17, 2026

Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The extract_zip function in sanitize.py employs zipfile.ZipFile.extractall(), which is vulnerable to the 'Zip Slip' path traversal attack. A maliciously crafted Office file (which is a ZIP archive) can contain entries with relative path sequences like ../../ to escape the temporary directory and overwrite sensitive files elsewhere on the filesystem.\n- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its processing of untrusted data from user-supplied Office documents.\n
  • Ingestion points: The sanitize.py script reads multiple XML files (document.xml, app.xml, core.xml, etc.) from within the extracted archive to perform fingerprint scanning and metadata updates.\n
  • Boundary markers: No markers or instructions are utilized to differentiate untrusted XML content from the script's logic.\n
  • Capability inventory: The script possesses extensive file system read/write capabilities and executes with the privileges of the user environment.\n
  • Sanitization: The script relies on basic regular expressions for manipulation and lacks structural validation or sanitization of the input XML before processing.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Jun 17, 2026, 12:43 PM
Security Audit — agent-trust-hub — deai-office