Guide · Agent skill
How to use the Statistical Research Orchestrator agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The Statistical Research Orchestrator is an agent skill that tells an AI agent to carry out a statistical research project in a fixed order: formulate the problem, propose a method, analyse its theory, run experiments, compare against baselines, write a report, and audit the chain. This guide explains what the skill does, how to install it, how to trigger it, and what it needs before you rely on it. It is written for readers who have never used agent skills.
Statistical Research Orchestratoraiming-lab/AutoResearchClaw · MIT Download the skill Quick start
- 1Download and unzip. Download the ZIP from this page and unzip it to get SKILL.md and its folder.
- 2Place the skill. Move the folder into ~/.claude/skills/ for personal use, or into .claude/skills/ in a project.
- 3Install the sub-skills. Add the seven referenced stat-* sub-skills, since they are not included in this file.
- 4Prepare your environment. Set up a working Python or similar environment for the experiment code.
- 5Describe your topic. Give a topic prompt, topic file or dataset description with your target and assumptions.
- 6Check each progress file. Read each stage's progress file before the next stage runs, starting with the problem formulation.
- 7Review the audit. Read the step 6 quality audit and treat any WARN or FAIL as a reason to check the work.
What is an agent skill, and what does this one do?
An agent skill is a folder containing a SKILL.md file that gives an AI agent instructions for a particular kind of task. When the skill is installed and a request matches it, the agent follows those instructions.
This skill is a coordinator and process guide. It sets the order of seven stages, each handled by a named sub-skill: stat-problem-formulator, stat-method-proposer, stat-theory-analyzer, stat-experiment-designer, stat-comparison-analyst, stat-result-synthesizer and stat-quality-auditor. Each stage writes a markdown progress file that the next stage reads. The agent is told to stop if the target or assumptions are undefined.
The skill does not contain statistical methods or code of its own. Any methods, proofs or code come from the referenced sub-skills.
- Formulate the problem: target, assumptions, hypotheses and evaluation criteria
- Propose a method with baselines, diagnostics and ablations
- Analyse theoretical properties and label what is proven, heuristic or only experimentally supported
- Design and run experiments that test theoretical predictions
- Compare results against baselines and check theory against experiment
- Write a final report and audit the whole chain
How do you install the stat-research-orchestrator skill?
Download the ZIP file from this page and unzip it. The skill is delivered as a single file, SKILL.md. To make it available to Claude Code, place the unzipped folder in your personal skills directory ~/.claude/skills/, or in a project's .claude/skills/ directory if you want it limited to one project.
To use it in the Claude apps instead, upload the skill under Settings → Capabilities → Skills.
Install the referenced sub-skills as well. They are not included in this file, and the orchestrator depends on them.
- Personal install: ~/.claude/skills/ (available across your projects)
- Project install: .claude/skills/ inside your project folder
- Claude apps: Settings → Capabilities → Skills, then upload the skill
What do you need before you start?
The skill needs the seven sub-skills listed above installed alongside it. The source does not include them, so check that each one is present before starting a project.
Experiment steps reference code and config files that the experiment designer creates. The source implies a working Python or similar environment but does not specify one, so you will need to set that up yourself.
The skill is licensed under MIT.
- All seven referenced sub-skills installed
- A working Python or similar environment for experiment code
- A multi-stage research project (see when not to use it below)
How do you trigger the skill with a natural request?
The skill is matched to requests about statistics research, problem formulation, method proposals, theory, experiments, comparison and results. Describe your research topic, and say what you want: a full research pipeline, a formal problem statement, or a comparison of methods.
Provide the topic source and any requirements at the start. The first stage needs a topic prompt, a topic file, or a dataset description to work from. Give the target you care about and your assumptions if you already have them, because the agent will stop if they are undefined.
- Example request: "Use the statistical research orchestrator to study whether a shrinkage estimator beats the sample mean under heavy-tailed noise."
- Example request: "Run the stat research pipeline on this topic file and give me a formal problem statement first."
- Example request: "Set up a theory-first study comparing two estimators, with baselines and an audit at the end."
What files does the pipeline produce?
Each stage writes a progress file under progress/<TOPIC_ID>/, and the next stage reads it. The experiment stage writes its config and results under experiments/<TOPIC_ID>/. The final audit is written to progress/<TOPIC_ID>/step6_quality_audit.md with a status of PASS, WARN or FAIL.
Check these files to see what the agent decided and why. The problem formulation file is the first place to look if the work seems to be going in the wrong direction.
- progress/<TOPIC_ID>/step0_problem_formulation.md
- progress/<TOPIC_ID>/step1_method_proposal.md
- progress/<TOPIC_ID>/step2_theory_analysis.md
- progress/<TOPIC_ID>/step3_experimental_evaluation.md
- progress/<TOPIC_ID>/step4_comparison.md
- progress/<TOPIC_ID>/step5_result_synthesis.md
- progress/<TOPIC_ID>/step6_quality_audit.md
experiments/<TOPIC_ID>/config.yamlWhat should you watch for when running it?
The skill runs scripts, because experiment steps create and run code. It does not use the network, according to the source. Review any scripts the experiment designer produces before running them on data you care about, and keep the work in a project folder you control.
Theory from the agent may be wrong or only heuristic. The pipeline asks the theory stage to label each result honestly, but you should check proof sketches with an expert before relying on them.
- Scripts run as part of the experiment stage, so review generated code first
- The source says the skill does not use the network
- Treat theory and proof sketches as drafts for expert review
What people use it for
- Structuring a statistics thesis with explicit assumptions and baselines
- Checking whether a proposed estimator beats simple baselines under stated conditions
- Building an automated research pipeline with audit checkpoints between stages
- Learning how a theory-first statistical study is organised from formulation to conclusions
Frequently asked questions
How do I install the stat-research-orchestrator skill?
Download the ZIP from this page and unzip it. Place the folder in ~/.claude/skills/ for personal use or in a project's .claude/skills/ folder, or upload it in the Claude apps under Settings → Capabilities → Skills.
Is the stat-research-orchestrator skill free?
The skill is licensed under MIT. The source does not state any price, so check the licence file for the terms that apply.
What does the stat-research-orchestrator need to run?
It requires the seven referenced sub-skills, which are not included in this file. The experiment stage implies a working Python or similar environment, which the source does not specify.
Does this skill do the statistics itself?
No. It is a coordinator and process guide. Statistical methods and code come from the referenced sub-skills.
When should I not use this skill?
Do not use it for quick data cleaning or charting, since the pipeline is heavy and slow. Also avoid relying on its output for high-stakes decisions without expert review.
Why does the agent stop partway through?
The agent is told to stop if the target or assumptions are undefined at the problem formulation stage. Define them in your request to avoid this.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you are a researcher, student or developer who wants a theory-first, staged statistical study with explicit assumptions, baselines and audit checkpoints. Pick something else for quick data cleaning, charting, or any decision that needs confident answers without expert review.
Download the skill