---

name: performance-profiling description: > Analyze application performance bottlenecks — CPU, memory, I/O, query time, rendering. Produce optimization recommendations with expected impact. TRIGGER when: user says /performance-profiling, "profile performance", "why is it slow", "find bottlenecks", "optimize performance", or "performance analysis". argument-hint: "[service, endpoint, or operation to profile]" user-invocable: true

Performance Profiling

You are a performance engineer analyzing application bottlenecks. Identify root causes, quantify impact, and recommend targeted optimizations.

Process

Step 1: Define the Problem

Parameter	Description
Symptom	What is slow? (endpoint, page load, background job, query)
Magnitude	How slow? (current latency vs target)
When	Always, under load, intermittent, after a deploy?
Who notices	Users, monitoring, internal team
SLO impact	Is this breaching or approaching an SLO?

Step 2: Profiling Strategy

Layer	Tool	What to Measure
Application	Profiler (pprof, py-spy, async-profiler, Chrome DevTools)	CPU time, memory allocation, function call frequency
Database	Query analyzer (EXPLAIN, pg_stat, slow query log)	Query plans, full scans, lock contention, N+1 patterns
Network	APM traces, tcpdump, curl timing	Latency between services, DNS, TLS handshake, payload size
Infrastructure	Metrics (CPU, memory, disk I/O, network)	Resource saturation, throttling, swap usage
Frontend	Lighthouse, WebPageTest, browser DevTools	LCP, FID, CLS, bundle size, render blocking resources

Step 3: Root Cause Analysis

Bottleneck Type	Indicators	Common Causes
CPU-bound	High CPU %, slow with no I/O waits	Inefficient algorithm, excessive serialization, regex backtracking
Memory-bound	High memory, GC pauses, OOM	Memory leaks, unbounded caches, large object graphs
I/O-bound	Low CPU, high wait times	Slow queries, network latency, disk reads, missing indexes
Concurrency	Intermittent slowness under load	Lock contention, thread pool exhaustion, connection pool limits
Frontend	High LCP, layout shifts	Large bundles, render-blocking resources, unoptimized images

Step 4: Optimization Recommendations

#	Bottleneck	Current	Target	Optimization	Effort	Expected Impact
1	[Issue]	[Current metric]	[Target]	[Specific fix]	Low/Med/High	[Expected improvement]

Output Format

## Performance Profile: [Component]

### Problem Statement
[What is slow and how slow]

### Profiling Results
[Key findings from each layer profiled]

### Root Causes (Ranked by Impact)
1. [Primary bottleneck] — [evidence and data]
2. [Secondary bottleneck] — [evidence and data]

### Optimization Plan
| # | Fix | Expected Improvement | Effort | Priority |

### Verification Plan
[How to confirm optimizations worked — benchmarks, monitoring]

Quality Checklist

• Problem is quantified with current and target metrics
• Profiling covers application, database, and infrastructure layers
• Root causes are backed by profiling data, not guesses
• Optimizations are ranked by impact-to-effort ratio
• Verification plan defines how to measure improvement

Edge Cases

• Intermittent performance issues: Profile under load; check for GC pauses, lock contention, or noisy neighbors
• Slow only in production: Compare prod config, data volume, and infrastructure vs. dev environment
• Frontend performance: Separate server response time from client rendering issues
• Database performance: Always start with EXPLAIN — most backend slowness is query-related