Experiment Design Skill for Claude
Design experiment plans with progressive stages — initial implementation, baseline tuning, creative research, and ablation studies. Plan baselines, datasets, hyperparameter sweeps, and evaluation metrics. Use when planning experiments for a research paper.
Install
one command, takes just this skill from the repositorynpx skills add https://github.com/lingzhi227/agent-research-skills --skill experiment-designThe instruction itself
13 sections, as written by the authorExperiment Design
Design structured, progressive experiment plans for research papers.
Input
$0— Research idea, plan, or method description
References
- 4-stage progressive experiment prompts:
~/.claude/skills/experiment-design/references/stage-prompts.md
Scripts
Generate experiment design
python ~/.claude/skills/experiment-design/scripts/design_experiments.py --plan research_plan.json --output experiment_design.json
python ~/.claude/skills/experiment-design/scripts/design_experiments.py --method "contrastive learning" --task classification --format markdown
Generates baselines, ablation matrix, hyperparameter grid, metric selection. Stdlib-only.
4-Stage Progressive Framework (from AI-Scientist-v2)
Stage 1: Initial Implementation
- Focus on getting a basic working implementation
- Use a simple dataset
- Aim for basic functional correctness
- Completion: at least one working (non-buggy) implementation
Stage 2: Baseline Tuning
- Tune hyperparameters (learning rate, epochs, batch size)
- Do NOT change model architecture
- Test on at least TWO datasets
- Completion: stable training curves, improvement over Stage 1
Stage 3: Creative Research
- Explore novel improvements and insights
- Be creative and think outside the box
- Test on at least THREE datasets
- Completion: demonstrated novel improvement
Stage 4: Ablation Studies
- Systematic component analysis
- Each ablation tests a different aspect
- Use same datasets as Stage 3
- Completion: all planned ablations done
Output Format
{
"stages": [
{
"name": "initial_implementation",
"goals": ["Basic working baseline", "Simple dataset"],
"max_iterations": 5,
"completion_criteria": "Working implementation with non-zero accuracy"
}
],
"baselines": ["Method A", "Method B"],
"datasets": ["Dataset1", "Dataset2", "Dataset3"],
"metrics": ["accuracy", "F1", "inference_time"],
"ablation_components": ["component_A", "component_B"],
"hyperparameter_grid": {
"lr": [1e-4, 1e-3, 1e-2],
"batch_size": [32, 64, 128]
},
"num_seeds": 3
}
Rules
- Always start simple (Stage 1) before complex experiments
- Each stage builds on the best result from the previous stage
- Multi-seed evaluation for statistical significance
- Document every experiment run in notes.txt
- Generate figures for training curves and comparisons
Related Skills
- Upstream: research-planning, idea-generation
- Downstream: experiment-code, data-analysis
- See also: paper-assembly
Other skills for the same job
different authors, same section of the catalogue
Use when the user is doing AI/ML work in a scientific domain such as biology, chemistry, physics, astronomy, climate, genomics, materials, medicine, ecology, energy, engineering, math, drug discovery, protein design, weather modeling, theorem proving, single-cell, or PDE solving. Hugging Science is a curated catalog of scientific datasets, models, blog posts, and interactive Spaces. This skill helps discover and use resources via `datasets`, `transformers`, the HF Inference API, `gradio_client`, and methodology citations.
Publish and manage research papers on Hugging Face Hub. Supports creating paper pages, linking papers to models/datasets, claiming authorship, and generating professional markdown-based research articles.
Semantic search, similar content discovery, and structured research using Exa API. Use when you need semantic/embeddings-based search, finding similar content, or searching by category (company, people, research papers, etc.).
Build RAG (Retrieval Augmented Generation) pipelines with web search and LLMs. Tools: Tavily Search, Exa Search, Exa Answer, Claude, GPT-4, Gemini via OpenRouter. Capabilities: research, fact-checking, grounded responses, knowledge retrieval. Use for: AI agents, research assistants, fact-checkers, knowledge bases. Triggers: rag, retrieval augmented generation, grounded ai, search and answer, research agent, fact checking, knowledge retrieval, ai research, search + llm, web grounded, perplexity alternative, ai with sources, citation, research pipeline
| Build RAG (Retrieval Augmented Generation) pipelines with web search and LLMs. research agent, fact checking, knowledge retrieval, ai research, search + llm, web grounded, perplexity alternative, ai with sources, citation, research pipeline
Web search and content extraction with Tavily and Exa via inference.sh CLI. Apps: Tavily Search, Tavily Extract, Exa Search, Exa Answer, Exa Extract. Capabilities: AI-powered search, content extraction, direct answers, research. Use for: research, RAG pipelines, fact-checking, content aggregation, agents. Triggers: web search, tavily, exa, search api, content extraction, research, internet search, ai search, search assistant, web scraping, rag, perplexity alternative
Web search and content extraction with Tavily and Exa via inference.sh CLI. Apps: Tavily Search, Tavily Extract, Exa Search, Exa Answer, Exa Extract. Capabilities: AI-powered search, content extraction, direct answers, research. Use for: research, RAG pipelines, fact-checking, content aggregation, agents. Triggers: web search, tavily, exa, search api, content extraction, research, internet search, ai search, search assistant, web scraping, rag, perplexity alternative
The protocol behind every investigation skill. Use when AI research must proceed without you: search-plan gate, Fact/Inference/Assumption labels, confidence stacking, diffable outputs.
How to use it
Take lingzhi227/experiment-design from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.