Lp3
Medium
- Category
- MCP Least Privilege
- Confidence
- 84% confidence
- Finding
- The skill explicitly instructs the agent to use network-capable functionality such as `gh api`, importing from GitHub, and serving/previewing content over HTTP, but the metadata declares no corresponding permissions. This creates a trust and governance gap: users and policy layers cannot accurately reason about the skill's external access, which can lead to unintended data exfiltration, unreviewed remote content ingestion, or policy bypass in environments that rely on declared permissions.
