Guard pipeline // Instruction-to-Action Trace
INSTRUCTION-TO-ACTION TRACE
See exactly how untrusted tool output or retrieved content influenced a proposed agent action — which spans the detectors flagged, the exact rule that blocked, held or allowed it, and the egress-destination policy verdict for the address it would have called. Every trace below is a real run through the same shield, IPI guard and policy proxy the product enforces with. “No detector hits” is reported as exactly that — never as “safe”.
Run a trace — free, no account
Honest scope: deterministic detector and policy rules with rule IDs. A trace shows what this pipeline actually decided on the input you gave it. ALLOW means the named rules did not block the action — it is not a safety guarantee.