threejs-object-sculptor
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a local Python helper script,
probe_reference_image.py, to retrieve technical metadata from reference images. This script is self-contained, uses only Python standard libraries, and implements manual binary parsing for common image formats (PNG, JPEG, GIF, WebP, BMP, TIFF) to safely extract dimensions and format information without full image decoding. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted image data. To mitigate risks associated with processing external content, it mandates a rigorous six-stage process—including inspection, decomposition, and verification—where assumptions must be explicitly labeled and anchors recorded. This structured approach helps ensure the agent's output remains bounded by technical evidence.
- [DATA_EXPOSURE]: File access is restricted to reading image headers for metadata extraction. The provided script reads a limited file prefix (default 1 MiB) and does not contain any logic for network communication or exfiltration of the data it processes.
Audit Metadata