Demo · real engine output
What it catches, and what it costs when nobody does.
Five things an agent does that you cannot see from the outside. Every finding below was produced by running the shipping detectors over the transcript shown — the wording, the severity and the quoted line are the engine’s, not ours.
What the agent did
If nobody is watching
Nothing here is a mock-up.
build-demo.py feeds each transcript to the same detectors that ship in the product and renders whatever comes back. A scenario the engine stops catching fails the build.