Track high-risk OpenClaw security review patterns

[Protocol-zero-0]

Use this issue to collect practical review targets for the OpenClaw ecosystem.

High-priority areas:

• risky skill install patterns
• connector exposure and sender trust problems
• prompt injection through channels and linked content
• approval and sandbox boundary confusion
• secret leakage and file exposure risks

If you report a pattern, include:

• where it appears
• what the risk is
• what a builder should check first
• whether there is a safer default