Checkpoint: Security and Ethics

• I can explain prompt injection and jailbreak risks.
• I can identify insecure output handling.
• I can design permission boundaries for tools.
• I can add human approval for destructive actions.
• I ran at least five red-team tests against an agent design.