performance-analysis

The performance-analysis skill acts as an expert performance engineer within the Agentic Startup framework. It is designed for developers and system architects who need to diagnose latency issues, resource saturation, or inefficient code paths. By strictly adhering to the methodology of 'measure first, optimize second,' this skill prevents premature optimization and ensures that changes are data-driven. It applies standard observability frameworks such as USE (Utilization, Saturation, Errors) for system resources and RED (Rate, Errors, Duration) for services to provide a comprehensive health assessment of the target codebase or infrastructure.

• Performs systematic baseline measurement to establish a ground truth before any modifications.

• Utilizes language-specific and infrastructure-level profiling tools to identify bottlenecks in CPU, memory, I/O, lock contention, or database queries.

• Evaluates performance data using p50, p95, and p99 percentiles to accurately represent tail latencies and user experience.

• Applies Amdahl's Law to prioritize optimizations, focusing efforts on the most impactful system components.

• Validates optimization results through secondary measurement to confirm actual performance gains against the established baseline.

• Use this skill when investigating performance regressions, planning for application scale, or optimizing critical execution paths.

• Always provide the analysis target as an argument; the agent will then guide you through profiling at Application, System, or Infrastructure levels.

• Follow the provided recommendation patterns—ranging from quick wins like caching and indexing to architectural changes like read replicas and horizontal scaling.

• Ensure production-like environments are used during profiling to obtain accurate telemetry data.

• Strictly avoid caching strategies without defined invalidation rules and ensure all recommendations cite empirical evidence from measurement output.