Keeping the Cursor app stable

This post details the implementation of systems for detecting and measuring application instability, specifically Out-of-Memory (OOM) errors. It describes dual debugging strategies: top-down investigation using feature flags and proxy metrics like oversize message payloads, and bottom-up investigation using crash stack analysis, heap snapshots, and continuous heap allocation profiling. Targeted mitigations for acute and slow-and-steady OOMs are discussed, along with the use of process isolation for extension stability. Prevention of regressions is addressed through Bugbot rules, agentic stress testing (skills), eliminating manual resource management, automated performance tests, and automated rollbacks on metric regressions.