perf

Guide structured performance investigations from baseline measurement to invasive profiling.

16|3|Updated May 5, 2026
One-click install
npx skills add https://github.com/Kevin-Liu-01/Agent-Machines --skill perf-kevin-liu-01
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: perf
Source: https://github.com/Kevin-Liu-01/Agent-Machines/tree/main/knowledge/skills/perf
Command: npx skills add https://github.com/Kevin-Liu-01/Agent-Machines --skill perf-kevin-liu-01

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Performance investigations are often chaotic and guess-based. This Skill provides a disciplined, data-driven approach that starts with inexpensive experiments and ramps up to more invasive tests, ensuring measurements guide every decision.

Core Features & Use Cases

  • Baseline measurement: three runs to report the median, establishing a trustworthy starting point.
  • Code-path inspection: end-to-end tracing to enumerate syscalls, locks, DB statements, and RPCs to identify bottlenecks.
  • Convergence ladder: learn from fast operations to slow ones and borrowing techniques from the lighter path.
  • Concurrency and scalability analysis: sweep parallelism (1, 4, 16, 64) to locate parallel ceilings and serial overhead.
  • Isolated probes: implement minimal standalone probes to validate hypotheses before production changes.
  • Production profiling tools: use perf/bpftrace when needed for deeper CPU attribution.

Quick Start

Start with a baseline measurement (three runs, report the median), then escalate through tracing, concurrency sweeps, and isolated probes before touching production.

Frequently Asked Questions about perf

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I identify and fix software performance bottlenecks using a data-driven approach?

To identify performance bottlenecks, establish a baseline measurement using three runs to report the median. Then escalate through end-to-end tracing, concurrency sweeps, and isolated probes before touching production code.

What is the best way to measure a reliable performance baseline before optimizing?

The best way to measure a reliable performance baseline is to run the target workload three times and report the median result. This establishes a trustworthy starting point for identifying true performance bottlenecks.

How do I locate parallel ceilings and serial overhead during concurrency analysis?

Locate parallel ceilings and serial overhead by sweeping parallelism across different concurrency levels, such as 1, 4, 16, and 64 threads. This scalability analysis reveals where parallel execution stops yielding improvements.

When do I need production profiling tools like perf or bpftrace for CPU attribution?

You need production profiling tools like perf or bpftrace for deeper CPU attribution when initial baseline measurements and end-to-end tracing fail to pinpoint the exact performance bottlenecks in your software.

How do I validate optimization hypotheses before applying changes to production?

Validate optimization hypotheses before production changes by implementing minimal standalone isolated probes. These reproducible experiments test specific bottlenecks without risking the stability of the main application environment.