Context-Inappropriate Capability
Medium
- Confidence
- 91% confidence
- Finding
- The skill introduces an external DeepSeek API path that sends user questions and patent background context to a third-party LLM, even though the skill is framed as building a local client demo portal. This creates a real data-handling risk because local/customer content may be transmitted outside the claimed local environment without strong purpose limitation, consent flow, or data-classification guidance.
