senior-data-scientist
by alirezarezvanisenior-data-scientist is a Claude skill for A/B test design, causal reasoning, feature engineering, and tabular ML evaluation. Use it to guide sample sizing, metric choice, leakage checks, SHAP review, and MLflow-style tracking; scripts are scaffold templates, not complete engines.
This skill scores 64/100, which means it is acceptable to list but should be presented as a broad, prompt-and-pattern skill rather than a production-ready data science toolkit. Directory users can understand when to invoke it and may get useful workflow scaffolding from SKILL.md, but the supporting scripts and references are too generic to justify strong confidence.
- Frontmatter has a detailed, triggerable description covering A/B testing, causal inference, predictive modeling, feature engineering, and business interpretation use cases.
- SKILL.md includes concrete workflow content and code snippets, including sample-size calculation and experiment analysis patterns, giving agents more structure than a generic data science prompt.
- Repository includes named references and scripts for experiment design, feature engineering, and model evaluation, signaling intended workflow coverage even if implementation depth is uneven.
- Support files appear mostly boilerplate: references repeat generic production principles rather than domain-specific statistical guidance.
- Scripts look scaffold-like, with placeholder comments such as "Add validation logic" and generic processing methods rather than complete experiment, feature engineering, or model evaluation tools.
Overview of senior-data-scientist skill
What senior-data-scientist is for
The senior-data-scientist skill is a GitHub-hosted Claude skill for statistical modeling, experiment design, causal inference, feature engineering, and predictive model evaluation. It is best suited for users who want an AI assistant to reason like a senior data scientist when planning A/B tests, reviewing tabular ML workflows, selecting evaluation metrics, or translating model results into business decisions.
Best-fit users and jobs
Use this skill when you already have a concrete analytics or Machine Learning problem and need structured help: sample size planning, two-proportion test interpretation, difference-in-differences framing, feature pipeline design, cross-validation strategy, AUC-ROC versus AUC-PR tradeoffs, SHAP-based explanation, or MLflow-style experiment tracking. The strongest fit is senior-data-scientist for Machine Learning on structured data, especially classification, regression, and controlled experiment analysis.
What makes it different from a generic prompt
A generic prompt may produce broad advice. The senior-data-scientist skill gives the agent a domain frame: experimental validity, statistical assumptions, feature leakage, evaluation design, and production-readiness. The upstream SKILL.md includes code-oriented examples for A/B testing and model workflows, while the support files suggest helper directions for experiment design, feature engineering, and model evaluation.
Important adoption cautions
This is not a complete plug-and-play data science package. The scripts/ files appear to be scaffold-style utilities rather than finished statistical engines, and the references/ files are high-level. Install it for agent guidance and workflow structure, not as a substitute for validating formulas, running your own notebooks, or reviewing statistical assumptions with your team.
How to Use senior-data-scientist skill
senior-data-scientist install and file review
Install from the repository with:
npx skills add alirezarezvani/claude-skills --skill senior-data-scientist
After install, read SKILL.md first because it contains the actual workflow substance. Then inspect scripts/experiment_designer.py, scripts/feature_engineering_pipeline.py, and scripts/model_evaluation_suite.py to understand the intended automation patterns. Treat references/experiment_design_frameworks.md, references/feature_engineering_patterns.md, and references/statistical_methods_advanced.md as orientation notes rather than authoritative documentation.
Inputs that make the skill useful
The skill performs best when you provide the decision context, not just the technique name. For experiments, include baseline rate, minimum detectable effect, traffic, assignment unit, primary metric, guardrail metrics, expected duration, and whether multiple comparisons are involved. For ML work, include target definition, dataset grain, leakage risks, class balance, train/test split constraints, business cost of false positives versus false negatives, and deployment environment.
Weak prompt: “Help me evaluate my model.”
Stronger prompt: “Use the senior-data-scientist skill to review a binary churn model. We have 1.2M customer-month rows, 7% positive rate, time-based split by signup month, XGBoost baseline, AUC-ROC 0.81, AUC-PR 0.29, and retention offers cost $40. Check leakage risks, metric choice, thresholding, calibration, and what to log in MLflow.”
Practical senior-data-scientist usage workflow
Start by asking the agent to restate assumptions before solving. Then request a plan, calculations or pseudocode, and a validation checklist. For an A/B test, have it separate design from analysis: sample size, randomization unit, eligibility, metric definition, power, then statistical test and interpretation. For feature engineering, ask for a pipeline that distinguishes raw fields, derived features, fit-only-on-training transformations, missing-value handling, and monitoring. For model selection, request cross-validation design, metric rationale, error analysis, and production monitoring.
When to use scripts versus prompts
Use the scripts as templates if you want to build local tooling with logging, config loading, and process structure. Do not assume they will perform complete experiment design, feature engineering, or model evaluation out of the box. For most users, the immediate value is invoking the skill in chat with rich context, then adapting any generated Python, SQL, or R code into your own tested environment.
senior-data-scientist skill FAQ
Is senior-data-scientist good for beginners?
It can help beginners learn the shape of professional data science work, but it assumes you can describe data, metrics, and modeling goals. If you are new to statistics, ask the agent to explain assumptions and failure modes in plain language before asking for code.
How is this different from asking Claude for data science help?
The skill narrows the assistant toward senior data scientist concerns: experimental design, causal validity, model evaluation, feature leakage, and business interpretation. That makes it more reliable for recurring analytics workflows than an open-ended prompt, but you still need to supply domain context and verify outputs.
Can it run full Machine Learning pipelines?
Not by itself. The repository contains scaffold-like Python scripts and workflow examples, not a complete AutoML or MLOps platform. Use the senior-data-scientist skill to design, critique, and generate components of a pipeline; run and validate those components in your own Python, R, SQL, Scikit-learn, XGBoost, or MLflow environment.
When should I not use this skill?
Avoid using it as the sole authority for regulated decisions, clinical analysis, financial risk models, or causal claims with weak identification. It is also a poor fit for unstructured deep learning work where the core task is computer vision, speech, or large-scale neural architecture tuning rather than tabular analytics and experiment design.
How to Improve senior-data-scientist skill
Improve senior-data-scientist outputs with better prompts
Give the skill the same information a senior reviewer would request: objective, data grain, time window, metric definitions, constraints, decision threshold, and what action will be taken from the result. Ask for “assumptions, risks, recommended method, code sketch, and validation checks” to prevent shallow answers.
Common failure modes to watch
Watch for metric mismatch, target leakage, underpowered experiments, post-treatment bias, multiple-testing inflation, inappropriate randomization units, and overclaiming causality from observational data. If the first answer skips these, explicitly ask the senior-data-scientist skill to audit the design for statistical and operational risks.
Iterate after the first answer
Do not stop at the first plan. Ask follow-ups such as: “What would invalidate this conclusion?”, “What sensitivity checks should I run?”, “Which metric should be the primary decision metric?”, “How would this change with a 3% baseline rate?”, or “Show the SQL/Python validation queries I should run before modeling.”
Strengthen the repository locally
If you adopt the skill heavily, improve it by adding project-specific templates: experiment intake forms, metric dictionaries, leakage checklists, model card formats, MLflow logging conventions, and tested utility scripts. The biggest upgrade is replacing generic scaffold code with validated functions for your team’s actual experiment, feature, and evaluation workflows.