Optimize Application Performance End-to-End

This skill orchestrates a comprehensive, five-phase application performance optimization workflow across the entire stack, dispatching specialized subagents via the Task tool for each step so that later phases build on data-driven findings from earlier ones rather than theoretical improvements. Phase 1 (Performance Profiling & Baseline) uses a performance-engineer subagent to generate CPU flame graphs, heap dumps, I/O traces, and database/API/frontend baseline metrics (with APM tools like DataDog or New Relic where available), an observability-engineer subagent to assess existing monitoring, OpenTelemetry tracing, and log aggregation for instrumentation gaps, and a further performance-engineer pass to measure Core Web Vitals (LCP, FID, CLS) and Real User Monitoring data. Phase 2 (Database & Backend Optimization) has a database-optimizer subagent tackle slow queries, missing indexes, Redis/Memcached caching, and connection pooling; a backend-architect subagent address N+1 queries, async patterns, pagination, response compression, and circuit breakers; and a performance-engineer subagent optimize service-to-service communication, message queues (Kafka/RabbitMQ), and distributed caching. Phase 3 (Frontend & CDN Optimization) has a frontend-developer subagent handle code splitting, tree shaking, lazy loading, and critical rendering path; a cloud-architect subagent configure CDN/edge caching (CloudFlare/CloudFront), image formats (WebP/AVIF), and HTTP/2/3 with Brotli compression; and a mobile-developer subagent optimize service workers, adaptive loading, and touch responsiveness for mobile/PWA. Phase 4 (Load Testing & Validation) runs comprehensive load tests with k6/Gatling/Artillery across normal, peak, and stress scenarios, then builds automated performance regression tests wired into CI/CD (Lighthouse CI, Artillery, GitHub Actions) with automatic rollback triggers. Phase 5 (Monitoring & Continuous Optimization) sets up production APM, distributed tracing, Grafana dashboards, PagerDuty alerts, and SLI/SLO definitions, then establishes an ongoing process of performance-budget tracking, A/B testing of performance changes, continuous production profiling, and regular review cycles. Configurable options include performance_focus (latency/throughput/cost/balanced), optimization_depth (quick-wins/comprehensive/enterprise), available tooling, budget constraints, and rollout tolerance (zero-downtime/maintenance-window/gradual). It defines concrete success criteria: P50 <200ms/P95 <1s/P99 <2s response times, Core Web Vitals thresholds (LCP <2.5s, FID <100ms, CLS <0.1), 2x peak load support at <1% error rate, database query P95 <100ms, CPU <70%/memory <80% utilization, at least 30% cost-per-performance improvement, and 100% critical-path monitoring coverage.

Use this skill when coordinating performance optimization across backend, frontend, and infrastructure together, establishing baselines and profiling to find real bottlenecks, designing load tests or capacity plans, or building performance/reliability observability. Do not use it for a small, localized fix with no broader performance goal, when there's no access to metrics, tracing, or profiling data to ground the work in, or for requests unrelated to performance or scalability. Load testing production requires approval and safeguards, and performance changes should roll out gradually with rollback plans.

Input is the target application/system ($ARGUMENTS) plus configuration for focus, depth, available tooling, and budget/rollout constraints. Output is, per phase: profiling reports and baselines, database/backend/distributed-system optimizations, frontend/CDN/mobile optimizations, load test results and CI/CD-integrated regression tests, and production monitoring dashboards, alerts, and a continuous optimization process.

Coordinates specialized subagents (performance-engineer, observability-engineer, database-optimizer, backend-architect, frontend-developer, cloud-architect, mobile-developer, test-automator) and references APM/observability tooling (DataDog, New Relic, Dynatrace, OpenTelemetry, Grafana, PagerDuty), load-testing tools (k6, Gatling, Artillery), and CDN/edge platforms (CloudFlare, CloudFront).

Teams and engineers coordinating an end-to-end, cross-stack performance optimization effort - backend, frontend, infrastructure, and monitoring together - who need a phased, data-driven process rather than isolated tuning of one layer.