blender-retarget
Pass
Audited by Gen Agent Trust Hub on Nov 20, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The script
scripts/brt_arp.pyusesimportlib.import_moduleto dynamically load modules from the user's Blender addons directory. This is done to locate and interface with the Auto-Rig Pro addon's internal helper functions by scanning for specific function definitions within the addon's source files. - [COMMAND_EXECUTION]: The test suite file
scripts/tests/test_audit_exit.pyutilizessubprocess.runto programmatically invoke the Blender executable. This is used to perform automated auditing and integrity checks on saved Blender files outside the primary process. - [INDIRECT_PROMPT_INJECTION]: The skill represents an ingestion surface for indirect instructions through the processing of external data sources including FBX files, JSON-based animation manifests, and NumPy
.npzbackup files. Although the skill employs safe loading configurations (such asallow_pickle=Falsefor NumPy), the ingestion of untrusted external assets is a documented risk factor. - Ingestion points: manifest files (
scripts/brt_library.py), FBX skeletons (scripts/brt_arp.py), and weight backup files (scripts/brt_weights.py). - Boundary markers: None identified; data is processed as standard structured formats (JSON/NPZ).
- Capability inventory: Broad filesystem access via
bpy, file read/write for backups, and subprocess execution in tests. - Sanitization: Uses
json.loadfor structured data and explicitly disables pickle loading for NumPy files to prevent arbitrary code execution during data restoration.
Audit Metadata