What problem does it solve? Tuning a paddle-defense simulation by hand is guesswork: you cannot tell whether a court is too easy, too hard, or produces a natural rally arc without a structured way to edit, validate, and observe it. This Skill gives you a repeatable loop for shaping pong.json, verifying it, and judging rally quality in the OpenHarness viewer. ## Core Features & Use Cases - Court configuration: Edit pong.json fields such as title, courtW/courtH, speed, maxSpeed, accel, and a style line that steers Jev's paddle strategy, with live updates in the viewer. - Schema validation: Run node "$JEV_DSH/toolchain/check.mjs" to reject invalid courts (missing title, out-of-range values like speed 1..60 or stepMs 30..2000) and print the pace at which the paddle gets outrun. - Direct model evaluation: Use toolchain/jev.mjs to ask Jev which paddle move keeps a rally alive for a given state, with a deterministic mock when TYPESAFE_API_KEY is unset. - Use Case: Raise speed and accel, re-run the checker, then watch the viewer to confirm rallies shorten and the miss count rises instead of the paddle never dropping the ball. ## Quick Start Ask the agent to tune pong.json so Jev holds a few returns before the accelerating ball slips past, then validate the court with the check script.
