autocad-automation

Pass

Audited by Gen Agent Trust Hub on Aug 10, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The script scripts/acad_preview.py invokes subprocess.run to execute a PowerShell command. This occurs when the Pillow library is missing and the script attempts to convert a BMP preview to PNG using the -ExecutionPolicy Bypass flag.
  • [COMMAND_EXECUTION]: The file scripts/acad_session.py utilizes the ctypes.windll.user32 library to manage the AutoCAD application window. It specifically calls ShowWindow and SetForegroundWindow to force the application into the foreground during automated drawing tasks.
  • [DATA_EXFILTRATION]: The skill handles file paths and external JSON inputs in scripts/acad_draw.py and scripts/acad_preview.py using Path().resolve() and direct property assignment without sanitization. These values are passed directly to AutoCAD COM methods for saving and opening drawings.
  • [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface due to how it processes external drawing briefs. Ingestion points: The agent ingests untrusted data from JSON files in scripts/acad_draw.py. Boundary markers: No explicit delimiters or instructions are provided to the agent to ignore instructions embedded within the JSON data. Capability inventory: The skill possesses the ability to execute shell commands via subprocess.run, manipulate the Windows GUI via ctypes, and modify the file system and CAD databases via the win32com.client COM interface. Sanitization: The skill lacks validation or escaping for drawing parameters (such as text content or layer names) provided in the JSON brief before they are executed within the AutoCAD environment.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 10, 2026, 03:05 AM
Security Audit — agent-trust-hub — autocad-automation