<skill_overview> <purpose>Improve runtime performance with allocation-aware and measurable changes</purpose> <triggers> <trigger>Profiling slow code paths</trigger> <trigger>Reducing allocations or GC pressure</trigger> <trigger>Optimizing async or I/O-heavy code</trigger> <trigger>Improving startup or throughput</trigger> </triggers> <sources> <source url="https://learn.microsoft.com/en-us/dotnet/standard/performance/">.NET Performance</source> <source url="https://learn.microsoft.com/en-us/dotnet/standard/memory-and-spans/">Memory and Spans</source> <source url="https://learn.microsoft.com/en-us/dotnet/standard/garbage-collection/fundamentals">GC Fundamentals</source> </sources> </skill_overview> <measurement> <rules> <rule>Measure before and after every change</rule> <rule>Use representative inputs and realistic data sizes</rule> <rule>Optimize the slowest 10% first</rule> </rules> </measurement> <allocations> <rules> <rule>Avoid unnecessary allocations in hot paths</rule> <rule>Reuse buffers with ArrayPool for large temporary arrays</rule> <rule>Use StringBuilder for repeated concatenation</rule> <rule>Prefer stackalloc + Span<T> for small, short-lived buffers</rule> </rules> <example> <code> Span<byte> buffer = stackalloc byte[128]; // Fill buffer without heap allocation </code> </example> </allocations> <spans_and_memory> <rules> <rule>Prefer Span<T> and ReadOnlySpan<T> for sync APIs</rule> <rule>Use Memory<T> only when data must outlive the stack frame</rule> <rule>Do not capture Span<T> across async/await</rule> </rules> </spans_and_memory> <linq_and_enumeration> <rules> <rule>Avoid multiple enumeration; materialize once if needed</rule> <rule>Prefer simple loops over complex LINQ in hot paths</rule> </rules> </linq_and_enumeration> <async_performance> <rules> <rule>Use ValueTask for cached or synchronous completions</rule> <rule>Avoid sync-over-async (.Result, .Wait)</rule> <rule>Pass CancellationToken to cancellable operations</rule> </rules> </async_performance> <gc_considerations> <rules> <rule>Keep large objects (>= 85 KB) to a minimum</rule> <rule>Dispose IDisposables promptly</rule> <rule>Avoid long-lived object graphs when possible</rule> </rules> </gc_considerations> <anti_patterns> <avoid name="premature_optimization">Optimizing without measurements</avoid> <avoid name="hot_path_linq">Heavy LINQ in tight loops</avoid> <avoid name="large_object_churn">Repeated LOH allocations</avoid> </anti_patterns>