CirvixAttack Lab
See the boundary
hold in under 60 seconds.
A hostile instruction can arrive through trusted content. CIRVIX does not claim to detect prompt injection itself; it evaluates the downstream action the agent attempts and records the result.
Scenario selector
Loading scenario
— Select a scenario to run.
audit event pendingProtect your own agent with the same local boundary.
Protect your own agent → Install FreeBrowser scenarios are deterministic walkthrough fixtures for comprehension. The authoritative engine runs locally. Reproduce the flagship sequence with npx @cirvix_ai/agent-control demo --fast and inspect the audit record with cirvix logs.