Bayesian Statistician Plugin

Claude Code plugin providing specialized subagents and skills for Bayesian statistical modeling workflows.

Quick Start

# Clone the repo
git clone https://github.com/xiandasun/bayesian-statistician-plugin.git

# Launch Claude Code with the plugin
claude --plugin-dir ./bayesian-statistician-plugin

Then just ask Claude to do Bayesian analysis - the bayesian-workflow skill auto-invokes:

> Analyze the dataset in data/sales.csv and build a Bayesian model

Installation Options

Option 1: Local directory (recommended for testing)

claude --plugin-dir /path/to/bayesian-statistician-plugin

Option 2: Install as plugin

/plugin marketplace add xiandasun/bayesian-statistician-plugin
/plugin install bayesian-statistician@xiandasun/bayesian-statistician-plugin

What's Included

Orchestration Skill

Skill	Description
bayesian-workflow	Full orchestration workflow (EDA → Model Design → Fitting → Validation → Reporting). Auto-invokes on Bayesian modeling tasks.

Subagents

Agent	Description
eda-analyst	Exploratory data analysis for Bayesian modeling
model-designer	Proposes candidate Bayesian models
model-fitter	Fits Stan models via CmdStanPy
model-critique	Assesses model fit and suggests improvements
model-refiner	Refines models based on critique
model-selector	Compares models via LOO/WAIC
prior-predictive-checker	Validates prior predictive distributions
posterior-predictive-checker	Validates posterior predictive checks
recovery-checker	Parameter recovery simulation
report-writer	Generates final analysis reports
decision-auditor	Audits model selection decisions

Technical Skills

Skill	Description
stan-coding	Best practices for writing Stan programs
visual-predictive-checks	Guidelines for predictive check visualizations
convergence-diagnostics	MCMC convergence diagnostic guidelines
artifact-guidelines	Standards for analysis artifacts
stan-ode-modeler	ODE modeling in Stan

Technical Stack

• Inference: Stan via CmdStanPy
• Diagnostics: ArviZ
• Package management: uv

Workflow Overview

The bayesian-workflow skill orchestrates a multi-phase analysis:

1. Phase 1: EDA → eda/ - Data exploration with parallel analysts
2. Phase 2: Model Design → experiments/experiment_plan.md - Propose competing models
3. Phase 3: Model Development → experiments/ - Validate, fit, critique, refine
4. Phase 6: Reporting → final_report.md - Generate final report

Each phase uses specialized subagents that communicate via files.