Vague Triggers
Medium
- Confidence
- 79% confidence
- Finding
- The activation examples are broad enough that an agent may invoke this skill for loosely related book-recommendation requests without clear user consent to perform external web lookups. In this skill, over-broad triggering is more dangerous because execution leads to live searches against third-party sites and retrieval of external reading links, expanding data exposure and increasing the chance of unintended browsing behavior.
