create-experiment

What problem does it solve?

This Skill eliminates the messy, untracked ad-hoc tinkering that occurs when testing technical hypotheses or comparing competing approaches, ensuring every exploration produces a reproducible, documented result instead of lost terminal history.

Core Features & Use Cases

• Structured Goal Refinement: Interactive workflow to align on clear success criteria and time boxes before starting any work, avoiding wasted effort on vague explorations.
• Isolated Sandbox Execution: Runs all tests in separate git worktrees or local experiment directories to avoid polluting your main codebase.
• Comprehensive Result Tracking: Captures all setup commands, execution outputs (including failures), generates comparison tables, and produces a dated markdown report with clear recommendations.
• Use Case Example: Use this Skill to test if ripgrep is faster than grep on your repo, validate if switching from pnpm to bun improves install times, or prototype a new feature without modifying your main codebase.

Quick Start

Ask the AI to run a scoped experiment to test if ripgrep outperforms grep on your current repository, with a 30 minute time box and success criteria of at least 20% faster search times for large codebases.