blender

Pass

Audited by Gen Agent Trust Hub on Apr 30, 2026

Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses shell commands including curl, pkill, cat, and python3 to manage the Blender application and communicate with its automation server.
  • [REMOTE_CODE_EXECUTION]: Core functionality involves sending arbitrary Python code to be executed within the Blender process via an HTTP interface on localhost:5656.
  • [PROMPT_INJECTION]: The skill provides an indirect prompt injection surface by translating natural language user requests into executable Python scripts.
  • Ingestion points: User instructions for Blender scene manipulation or automation (SKILL.md).
  • Boundary markers: Uses quoted heredoc delimiters (<<'PYEOF') to prevent shell-level interpolation during script execution.
  • Capability inventory: Full access to the Blender Python API (bpy), file system operations via the OUTPUT path, and subprocess management through shell utilities.
  • Sanitization: No explicit validation or sandboxing of the generated Python code is performed before execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 30, 2026, 10:59 AM
Security Audit — agent-trust-hub — blender