agents in

skills/software/performance-optimization.md

@@ -0,0 +1,45 @@
+# Performance Optimization
+
+## Purpose
+
+Improve responsiveness and efficiency by focusing on the bottlenecks that matter most to users, systems, or operating cost.
+
+## When to use
+
+- Investigating slow pages, endpoints, jobs, or queries
+- Reducing memory, CPU, network, or rendering overhead
+- Preventing regressions in critical paths
+- Prioritizing optimization work with limited time
+
+## Inputs to gather
+
+- Performance symptoms, target metrics, and critical user or system paths
+- Existing measurements, profiles, logs, traces, or benchmarks
+- Current architecture and known hot spots
+- Acceptable tradeoffs in complexity, cost, and feature scope
+
+## How to work
+
+- Measure or inspect evidence before optimizing.
+- Focus on the dominant bottleneck rather than broad cleanup.
+- Prefer changes that improve the critical path without making the system harder to maintain.
+- Re-measure after changes when possible.
+- Capture the conditions under which the optimization matters so future work does not cargo-cult it.
+
+## Output expectations
+
+- Bottleneck diagnosis and recommended or implemented improvement
+- Before-and-after evidence when available
+- Notes on tradeoffs, limits, and remaining hot spots
+
+## Quality checklist
+
+- Optimization targets a real bottleneck.
+- Claimed gains are grounded in evidence, not assumption alone.
+- Complexity added by the optimization is justified.
+- Regression risk is considered for correctness and maintainability.
+
+## Handoff notes
+
+- Note whether the result is measured, estimated, or hypothesis-driven.
+- Pair with observability and operability when instrumentation is weak.