Agentic Experiment Designer — Testing skill for Claude Code

Autonomous Experiment Designer: Turn Any AI Agent Into a Self-Directed Research Scientist

[](https://hysohail.github.io/agentic-experiment-designer/)

**Tired of AI assistants that only answer questions?** Meet **Autonomous Experiment Designer** – the single-file framework that transforms any large language model (OpenAI, Claude, or local) into a proactive research assistant that *designs, tests, iterates, and validates experiments* without human hand-holding. Inspired by the vision of ResearcherSkill, this tool flips the script: instead of telling your AI *what* to do, you tell it *what to discover*.

---

🧪 What Makes This Different?

Traditional AI coding agents wait for instructions. **Autonomous Experiment Designer** gives your AI a scientific mindset. Think of it as giving a brilliant but passive librarian a fully stocked laboratory – suddenly they’re not just finding books, they’re mixing chemicals, running simulations, and publishing results.

**A single Python file** (`experiment_designer.py`) contains:

• A self-reinforcing hypothesis generator
• Automated experiment orchestration
• Real-time result analysis and refinement loops
• Multi-model support (OpenAI GPT-4, Claude 3, local HuggingFace models)

---

🧬 How It Works (The Research Loop)

graph TD
    A[User Provides Research Goal] --> B[Hypothesis Generator]
    B --> C{Multiple Hypotheses?}
    C -->|Yes| D[Experiment Designer]
    C -->|No| B
    D --> E[Test Environment Simulator]
    E --> F[Data Collector]
    F --> G[Analysis Engine]
    G --> H{Statistically Significant?}
    H -->|No| I[Refine Hypothesis]
    I --> B
    H -->|Yes| J[Generate Research Report]
    J --> K[Suggest Next Experiments]
    K --> B

The system doesn't stop until it either confirms a hypothesis or exhausts reasonable variations. Every cycle produces:

• A falsifiable prediction
• A controlled exp