Aborting Async Work Isn't Enough: Lifecycles and Output Contracts in a Tauri Toolbox

A developer walks through a problem where a user starts one conversion, switches tools, and starts another, so the earlier promise may resolve last and overwrite the newer screen. The write-up argues that such bugs reflect deeper architectural choices about who owns asynchronous work and when resources should be released. It separates three lifetimes in a desktop toolbox: the application, individual tool instances, and active operations, which do not necessarily end together. To manage cleanup, the project uses a scope object that tracks listeners, timers, abort controllers and revision tokens, and runs cleanup callbacks in reverse order. The author stresses that an abort signal alone is insufficient, because already-completed results can still reach a continuation, so results also need identity checks. The piece is a self-described AI-assisted teaching walkthrough of one source-available project, not a benchmark or security audit.