data-science-plugin

Data Science Plugin

Data science and ML workflow tools that compound institutional knowledge. 9 agents, 8 commands, 19 skills for problem framing, preprocessing, validation, EDA, experimentation, review, deployment, and knowledge compounding.

Install

Add the repo as a marketplace, then install:

/plugin marketplace add andikarachman/data-science-plugin
/plugin install ds@data-science-plugin

Prerequisites

The plugin's skills and agents use Python libraries for data analysis. Install them into your active environment:

uv pip install pandas scikit-learn scipy statsmodels numpy

Optional libraries (visualization, advanced models, and high-performance DataFrames):

uv pip install matplotlib seaborn aeon xgboost lightgbm shap polars optuna

Run /ds:setup to check which libraries are installed.

Workflow

Frame -> Preprocess -> Validate -> Explore -> Experiment -> Review -> Ship -> Compound -> Repeat

Command	Purpose
/ds:plan	Frame business questions as DS problems and plan approach
/ds:preprocess	Clean, validate, and transform raw data with automated pipelines
/ds:validate	Run formal data quality validation with expectation suites
/ds:eda	Run structured exploratory data analysis
/ds:experiment	Design and run rigorous ML experiments
/ds:review	Peer review experiments for methodology and reproducibility
/ds:ship	Assess deployment readiness and generate model cards
/ds:compound	Capture learnings to make future projects faster

Each cycle compounds: experiment learnings surface in future plans, error patterns inform feature engineering, and review feedback becomes institutional knowledge.

Components

Component	Count
Agents	9
Commands	8
Skills	19
Templates	9
MCP Servers	1

Agents

Analysis (4)

Agent	Description
problem-framer	Frame business questions as structured DS problems
data-profiler	Profile datasets for quality, structure, and anomalies
feature-engineer	Design and evaluate feature transformations
pipeline-builder	Assess raw data quality and design preprocessing pipelines

Modeling (2)

Agent	Description
experiment-designer	Design rigorous experiments with hypotheses and evaluation plans
model-evaluator	Evaluate performance with slicing, calibration, and fairness checks

Review (3)

Agent	Description
documentation-synthesizer	Synthesize findings into reusable learning documents
reproducibility-auditor	Audit experiments for reproducibility (seeds, versions, data hashes)
deployment-readiness	Evaluate models for production deployment readiness

Commands

Command	Description
/ds:plan	Search past learnings, frame the problem, plan the approach, output a plan doc
/ds:preprocess	Assess data quality, design and execute preprocessing pipelines, output preprocessing report
/ds:validate	Run data quality validation with Great Expectations, pandas, or data contracts, output validation report
/ds:eda	Profile data, analyze distributions, check quality, output an EDA report
/ds:experiment	Formulate hypothesis, design methodology, check for leakage, output experiment plan and results
/ds:review	Peer review experiments for methodology, leakage, reproducibility, and statistical validity
/ds:ship	Assess deployment readiness, generate model card and deployment documentation
/ds:compound	Extract learnings from completed work, categorize, and save to docs/ds/learnings/

Skills