office-repair

Warn

Audited by Gen Agent Trust Hub on Aug 6, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The archive_generator function in scripts/office_repair.py performs code injection by prepending a guard block to Python scripts. This injected code executes logic to prevent file overwrites and uses atexit handlers to restore data.
  • [COMMAND_EXECUTION]: The script scripts/test_office_repair.py invokes subprocess.run to execute Python scripts, including those modified by the archiving feature.
  • [COMMAND_EXECUTION]: The skill uses Windows COM and PowerShell (via the ane_package.officeops library) for Office document manipulation.
  • [PROMPT_INJECTION]: The skill processes untrusted content from Office documents through extract_text in scripts/office_repair.py, creating an indirect prompt injection vector.
  • Ingestion points: Content is read from Word, Excel, and PowerPoint files.
  • Boundary markers: No delimiters are used to separate ingested content from instructions.
  • Capability inventory: File writing, subprocess execution, and COM-based application control.
  • Sanitization: No sanitization is performed on extracted text to prevent instruction follow-through.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 6, 2026, 10:53 AM
Security Audit — agent-trust-hub — office-repair