Developer builds tool to visualize how AI instruction rules degrade under context load

A developer created an interactive browser-based demo to illustrate how AI agent instructions lose effectiveness as a session's context window fills up. The tool renders each instruction as a 3D hill, with height determined by writing quality, while a rising water level simulates increasing context load — weaker rules sink first. The project was built as a single self-contained HTML file with no external dependencies, using a hand-rolled 2D canvas renderer instead of WebGL. A companion linting tool analyzes instruction text for common weaknesses such as hedging, vague language, and poor directive ordering. The core argument is that critical prohibitions should be enforced via runtime code hooks rather than relying on a model's attention to prose instructions.