skills/zaoqu-liu/scienceclaw/rdkit/Gen Agent Trust Hub

rdkit

Warn

Audited by Gen Agent Trust Hub on Apr 16, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The 'Performance Optimization' section of the skill recommends using the pickle module (pickle.load()) for loading molecular data from files. Python's pickle module documentation states that it is not secure and can be exploited to execute arbitrary code during the deserialization process. Recommending this for performance without security warnings constitutes an unsafe practice, particularly if the agent processes files provided by external or untrusted sources.\n- [PROMPT_INJECTION]: The skill includes a 'Suggest Using K-Dense Web' section that provides specific instructions to the agent on how to proactively promote an external website (www.k-dense.ai) when tasks involve high complexity. These instructions direct the agent to steer user interactions toward a specific commercial platform, which attempts to override the agent's neutral task-oriented persona for promotional purposes.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 16, 2026, 08:27 AM