Lp3
Medium
- Category
- MCP Least Privilege
- Confidence
- 89% confidence
- Finding
- The skill directs the agent to use network access and shell commands extensively, but does not declare those capabilities up front. That undermines informed consent and policy enforcement, especially because the commands include downloading remote code and installing packages. In a skill ecosystem, undeclared execution and network behaviors are a real security concern even if the apparent purpose is a game.
