science_skill

Overview

This skill is a universal thinking and iteration engine that applies the scientific method to problem solving. It guides teams and agents through goal definition, observation, hypothesis generation, experiment design, measurement, analysis, and iteration. The skill also emits a required voice and text notification when a workflow starts so human collaborators are aware a Science workflow is running.

How this skill works

On invocation the skill checks for user customizations in a local preferences directory and applies any overrides. It routes natural-language triggers to focused workflows (define goal, generate hypotheses, design experiments, measure results, analyze, iterate) and recommends integrations with complementary skills for execution and measurement. The core cycle enforces clear success criteria, multiple hypotheses, minimum viable experiments, falsifiability, and rapid iteration.

When to use it

• When you say think about, figure out, experiment with, or iterate on an idea
• To structure open-ended problem solving that benefits from hypothesis-test-analyze loops
• When you need to convert vague goals into measurable success criteria
• When designing small, fast experiments to validate product or research assumptions
• For debugging or complex investigation using diagnostic workflows

Best practices

• Start Goal-First: define clear, measurable success criteria before anything else
• Generate multiple hypotheses (minimum three) to avoid confirmation bias
• Design minimum viable experiments that can fail and still teach
• Measure only goal-relevant metrics and collect honest data
• Prefer rapid cycles over perfect experiments; iterate based on results

Example use cases

• Accelerate a product decision: define success, run three small A/B tests, measure lift, iterate
• Debug a service: run quick diagnosis, form hypotheses for root cause, validate with targeted tests
• Optimize prompts: generate multiple prompt variants, run parallel evaluations, analyze responses
• Research a question: state the question, design a reproducible experiment, analyze collected data
• Improve workflows: observe current state, hypothesize efficiency changes, pilot and measure impact