Repair File Watcher, Hot Reload & Reload Leaks

The likely issue is a contract mismatch around file watchers, hot reload, duplicate subscriptions, recursive watches, stale handles, rebuild loops, configuration reloads, shutdown cleanup, path churn, and cross-platform watcher semantics. I would first define the invariant: Each logical watch target has one bounded owner and one teardown path; reloads may replace state but must not accumulate duplicate watchers, callbacks, or rebuild loops. Then I would trace the smallest owning boundary, apply or propose a bounded repair/review step, and add deterministic normal, boundary, failure and replay tests. The workflow preserves existing authorization, privacy, security policy, public interfaces, user-visible behavior, and documented operational ownership. If repository/workspace files are available, findings can be mapped to exact paths and focused local validation; with partial evidence, the result stays an explicit diagnosis/evidence map with assumptions and no invented execution claims.