benchmark

Benchmark Claude Code rules and skills with before/after comparisons.

Updated Oct 28, 2022
One-click install
npx skills add https://github.com/patrykk21/dotfiles --skill benchmark-patrykk21
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: benchmark
Source: https://github.com/patrykk21/dotfiles/tree/main/claude/skills/benchmark
Command: npx skills add https://github.com/patrykk21/dotfiles --skill benchmark-patrykk21

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires python, pypdf, pdfplumber, pdf2image, and includes scripts (resource) and references (resource) components.

What problem does it solve?

This Skill allows users to measure the impact of Claude Code rules, skills, and workflows by running quantitative before/after comparisons and evaluating the outputs against assertions.

Core Features & Use Cases

  • Benchmark and Evaluate: Measure the impact of rules, skills, and workflows through quantitative comparisons.
  • Assertion-Based Evaluation: Grade outputs against falsifiable, discriminative assertions.
  • Isolation and Robustness: Ensures filesystem sandboxing and global rule auto-hiding for accurate evaluation.
  • Customizable Workflows: Tailor benchmarking procedures to different target types and scenarios.
  • Data Analysis and Reporting: Generate detailed reports with pass rates, timing, and evidence of evaluation.

Quick Start

Run a benchmark to compare the performance of a rule pack with 'claude benchmark <rule-pack-name> --with rules <rule-pack-path> --without rules ~/.claude/rules/'.

Frequently Asked Questions about benchmark

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

FAQPage Schema
How do I evaluate the impact of Claude Code rules and skills quantitatively?

You can measure their impact by running quantitative before/after comparisons and evaluating outputs against falsifiable assertions to ensure accurate assessment.

What is assertion-based evaluation for Claude Code workflows?

Assertion-based evaluation grades Claude Code outputs against falsifiable, discriminative assertions to generate detailed reports with pass rates and timing.

How do I benchmark a rule pack in Claude Code?

Run a benchmark command comparing performance with and without the target rules, such as 'claude benchmark <rule-pack-name> --with rules <rule-pack-path> --without rules ~/.claude/rules/'.

Do I need Python to run Claude Code skill benchmarks?

Yes, you need Python installed along with pypdf, pdfplumber, and pdf2image to handle the filesystem operations and text extraction required for benchmarking.

How does benchmarking ensure accurate evaluation of Claude Code rules?

Benchmarking ensures accurate evaluation by enforcing filesystem sandboxing and auto-hiding global rules, isolating the target rules for precise measurement.

What reports are generated when measuring Claude Code workflows?

The benchmarking process generates detailed reports containing pass rates, timing metrics, and concrete evidence of the evaluation against assertions.