Intent-Code Divergence
Medium
- Confidence
- 95% confidence
- Finding
- The documentation asserts that the agent will request confirmation before calling non-example tool URLs, but the skill contains no implementation or enforceable control for such approval. This creates a misleading safety guarantee that could cause users or orchestrators to trust unverified outbound requests, enabling unintended SSRF-like fetches, privacy leaks, or access to sensitive internal endpoints if the backend honors arbitrary URLs.
