Lp3
Medium
- Category
- MCP Least Privilege
- Confidence
- 91% confidence
- Finding
- The skill instructs the agent to use shell commands and a Python script, and references writing an output report, but the skill metadata declares no permissions. This mismatch is dangerous because an agent or runner may execute capabilities the user did not explicitly approve, including repository inspection, local filesystem access, and file creation/modification during analysis workflows.
