How are the results formatted for analysis?

Metrics are emitted in a standardized JSON block between PERF_METRICS_START and PERF_METRICS_END markers, making them easy to parse by automated tools.

What is the minimum duration for a benchmark run?

The skill enforces a 60-second minimum for standard runs to ensure statistical significance, though it allows 30 seconds specifically for binary search phases.

Does the skill handle system warmups?

Yes, it requires a minimum 10-second warmup period before measurement begins to ensure the environment, such as the JIT compiler and caches, is stabilized.

Why are benchmarks forced to be sequential?

Sequential execution prevents resource contention between tests, ensuring that metrics like CPU usage and latency are accurate and not skewed by parallel background processes.

Can I modify code while a benchmark is running?

No, the skill includes a constraint that prohibits changing code during the benchmarking process to maintain the integrity of the results.

Performance Benchmarker

Name: Performance Benchmarker
Author: agent-sh

byagent-sh

•

Analytics & Monitoring

Conducts rigorous, evidence-backed performance investigations and baseline measurements with strict sequential execution rules.

The perf-benchmarker skill empowers Claude to perform high-integrity performance analysis by enforcing professional benchmarking standards. It eliminates common measurement errors by requiring sequential runs, mandatory 10-second warmups, and minimum 60-second durations to ensure statistical significance. Ideal for identifying regressions and establishing reliable baselines, the skill automatically handles anomalies and formats results into a standardized JSON schema for precise downstream analysis and reporting.

Key Features

01Built-in logic for identifying and re-running anomalous data points

02Implements mandatory 60-second minimum run durations for stability

03Enforces sequential execution to prevent resource contention and noise

041 GitHub stars

05Automated 10-second warmup periods to stabilize caches and JIT

06Standardized JSON metrics output with start/end markers

Use Cases

01Establishing high-confidence baselines for system throughput and latency

02Investigating non-deterministic performance anomalies through repeated measurement

03Validating performance regressions after major code changes

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add agent-sh/perf perf-benchmarker

For use in Claude.ai and ChatGPT

Download Skill