Skills
skills/internscience/chemclaw/mol-3d-viewer/Gen Agent Trust Hub

mol-3d-viewer

Pass

Audited by Gen Agent Trust Hub on Mar 30, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface when processing untrusted chemical data.\n
  • Ingestion points: The smiles and name parameters in scripts/mol_3d_viewer.py, which accept input from user commands or external files.\n
  • Boundary markers: No delimiters or explicit safety instructions are used to isolate user-provided chemical data within the generated HTML report structure.\n
  • Capability inventory: The skill generates .html files (interactive viewers), .png images, and .sdf structure files using Path.write_text and standard file writing libraries.\n
  • Sanitization: User-provided input, such as the chemical name, is interpolated directly into the HTML template in the generate_html method without escaping. This could allow for Cross-Site Scripting (XSS) if the generated file is opened in a browser after processing a specially crafted malicious name.\n- [EXTERNAL_DOWNLOADS]: Fetches chemical name-to-SMILES translations from the University of Cambridge's OPSIN service (opsin.ch.cam.ac.uk).\n- [EXTERNAL_DOWNLOADS]: The interactive HTML reports produced by the skill load the 3Dmol.js rendering library from a public CDN (3Dmol.org) to support WebGL visualization.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 30, 2026, 06:44 AM