A coding agent at Halden Pay, a fictional payments company, is moving its billing code to decimal arithmetic. What the agent tries is scripted. Whether it may is decided live, step by step, by the same engine Plumb runs. Nothing on the right is canned.
Start with what should happen most of the time: the agent reads code and config, edits, tests and pushes, and Plumb never gets in the way.
Enforce: Plumb refuses what it catches.
Task: …
Press Play. Each decision the engine makes appears here as the agent works.
The agent's choices. A stand-in for the model answers each turn with the next step of the scene, so every visitor sees the same story. No command is actually executed; each step's output is part of the script.
Everything Plumb does. The agent talks to the Enlil engine exactly as Claude Code does with Plumb installed: model traffic through the proxy, every action through the pre-run hook. The verdicts, reasons, traces and timings on this page come from that engine as the scene plays.
Every scene is also an automated test, run against the engine on each change, including the ones it misses. If the engine's behaviour changes, the test fails before this page can show something untrue.