cadsmith-text-to-cad

Warn

Audited by Gen Agent Trust Hub on Nov 7, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The execute function in scripts/cadsmith.py dynamically writes a temporary Python runner script and uses it to execute CAD scripts generated by the agent via runpy.run_path. Additionally, the log function contains a dynamic module loading pattern using __import__ for the hashlib library.
  • [COMMAND_EXECUTION]: The skill uses subprocess.run to execute Python scripts in a separate process. Although it utilizes the -I flag for isolated execution, this method does not implement a secure sandbox for the filesystem or network, potentially allowing generated code to interact with the host system.
  • [INDIRECT_PROMPT_INJECTION]: The skill's multi-agent workflow converts natural language prompts into executable Python code (part.py) which is then run automatically. This creates a significant surface for indirect prompt injection attacks.
  • Ingestion points: User-supplied text descriptions are ingested by the Planner and Coder agents to generate geometric plans and CAD scripts as defined in references/agents.md.
  • Boundary markers: The instructions do not specify the use of strict boundary markers or delimiters to isolate user-provided data from the agent's instructions.
  • Capability inventory: The skill has the capability to write files (open().write()) and execute shell commands via the execute tool in scripts/cadsmith.py.
  • Sanitization: There is no evidence of automated code sanitization or safety validation of the generated Python scripts prior to their execution in the host environment.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 7, 2026, 06:03 AM