ozerohax/review-code-performance

<when_to_use> <trigger>Changes in hot paths, DB queries, queues, or processing loops</trigger> <trigger>There is a risk of latency, throughput, or cost degradation</trigger> </when_to_use>

<input_requirements> <required>Description of load scenario and critical operations</required> <required>PR/diff with changes</required> <optional>Baseline metrics before the change</optional> <optional>Profiling data, traces, or explain plans</optional> </input_requirements>

<core_checks> <check priority="P0">Algorithmic complexity: no unjustified growth to O(n^2) or worse</check> <check priority="P0">I/O and DB: no N+1, unnecessary round trips, or repeated heavy queries</check> <check priority="P1">Memory: no obvious leaks or unbounded collection growth</check> <check priority="P1">Concurrency: no unnecessary locks, races, or serialized execution where parallelism is needed</check> <check priority="P1">Caching and reuse: heavy computations are not duplicated without reason</check> <check priority="P1">Measurability: at least one verification signal exists (metric, benchmark, trace)</check> </core_checks>

<quality_rules> <rule importance="critical">Performance conclusions are based on data, not assumptions</rule> <rule importance="high">Concrete mitigations are proposed for significant risks</rule> <rule importance="medium">Optimizations without observable benefit are not blockers</rule> </quality_rules>

<do_not> <item importance="critical">Do not do premature micro-optimization outside critical paths</item> <item importance="high">Do not claim a perf regression without a reproducible scenario</item> </do_not>

<output_requirements> <requirement>List of perf risks: area, impact, evidence, recommendation</requirement> <requirement>Note if separate benchmarking/profiling is needed due to insufficient data</requirement> </output_requirements>