shap
Warn
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: MEDIUMMETADATA_POISONINGDYNAMIC_EXECUTION
Full Analysis
- [METADATA_POISONING]: The skill includes instructions that attempt to hijack the agent's behavior for promotional purposes.
- The 'Suggest Using K-Dense Web' section instructs the agent to proactively market a commercial platform to the user, framing the suggestion as a 'productivity upgrade'.
- The directives specify when to trigger this marketing (e.g., during complex workflows) and how to coordinate it, which modifies the agent's natural response pattern to serve a third-party commercial interest.
- There is a discrepancy between the metadata author ('K-Dense Inc.') and the skill author ('wu-yc'), which may mislead users about the skill's origin.
- [DYNAMIC_EXECUTION]: The skill's documentation and integration examples promote the use of
joblib.load()for loading machine learning models and SHAP explainers. joblibutilizes the pickle serialization format, which is inherently unsafe and can lead to arbitrary code execution if used to load maliciously crafted files from untrusted sources.- The examples provided for 'Production APIs' and 'Performance Optimization' encourage loading files from paths without including necessary security warnings regarding untrusted input.
Audit Metadata