Python Analytics Skills

A plugin for Claude Code and other AI coding platforms providing Agent Skills for Bayesian modeling and reactive Python notebooks. Packages specialized knowledge for PyMC and marimo into skills that Claude loads on demand.

Skills

Skill	Description
pymc-modeling	Bayesian statistical modeling with PyMC 6+, PyTensor 3+, and ArviZ 1.1+. Covers model specification, inference, diagnostics, hierarchical models, GLMs, GPs/HSGPs, BART, time series, mixtures, causal models, priors, and custom likelihoods.
prior-elicitation	Prior selection, prior predictive checks, constrained priors, PreliZ workflows, expert priors, weakly informative priors, and prior sensitivity analysis.
pymc-extras	pymc-extras guidance for splines, distributional regression, R2D2/horseshoe priors, marginalization, and Laplace approximation.
pymc-testing	Testing PyMC models with pytest. Covers pymc.testing.mock_sample, fixtures, structure-only tests, and slow posterior inference tests.
model-evaluation	Bayesian model comparison and predictive evaluation with ArviZ 1.1 LOO/ELPD, stacking, model averaging, Bayes factors, and Pareto-k diagnostics.
marimo-notebook	Reactive Python notebooks with marimo. Covers CLI usage, @app.cell notebooks, UI components, layout, SQL integration, caching, state management, notebook conversion, templates, and wigglystuff widgets.

Installation

Via npx (Recommended — works across agents)

npx skills add pymc-labs/python-analytics-skills

One command, works with Claude Code, Cursor, Gemini CLI, and 15+ other agents.

As a Claude Code Plugin

Two-step process using Claude Code slash commands:

/plugin marketplace add pymc-labs/python-analytics-skills
/plugin install analytics@pymc-labs-python-analytics-skills

Installs all skills plus the keyword-suggestion hook. Supports /plugin update for future updates.

For Pi / Oh My Pi

Oh My Pi can install this repository as a marketplace plugin using the Claude-compatible .claude-plugin/marketplace.json catalog:

/marketplace add pymc-labs/python-analytics-skills
/marketplace install analytics@python-analytics-skills

CLI equivalent:

omp plugin marketplace add pymc-labs/python-analytics-skills
omp plugin install analytics@python-analytics-skills

For a project-local install, add --scope project to the install command.

Manual Installation

git clone https://github.com/pymc-labs/python-analytics-skills.git
cd python-analytics-skills
./install.sh claude                    # Claude Code
./install.sh all                       # All platforms
./install.sh claude -- pymc-modeling   # Specific skill only

Utility Commands

# List available skills with descriptions
./install.sh --list

# Validate skill structure
./install.sh --validate
./scripts/validate-skills.sh

Benchmark

The repo includes a PyMC skill benchmark in benchmark/pymc-modeling/. It compares Claude outputs with and without skills/pymc-modeling/SKILL.md injected via --append-system-prompt.

Benchmark contents:

• tasks.yaml — five Bayesian modeling tasks
• src/ — runner, scorer, analysis, and CLI modules
• tests/ — pytest coverage for the benchmark harness
• data/ — small input datasets used by tasks
• scripts/prepare_data.py — deterministic data preparation
• pixi.toml / pixi.lock — benchmark environment

Run benchmark commands from the benchmark directory:

cd benchmark/pymc-modeling
pixi run validate
pixi run test
pixi run run-all
pixi run score-all
pixi run analyze

Generated benchmark outputs such as results/, .nc files, figures, and archived runs are not package source.

Platform Support

Platform	Install Location	Auto-Discovered
Claude Code	~/.claude/skills/	Yes
OpenCode	~/.config/opencode/skills/	Yes
Gemini CLI	~/.gemini/skills/	Yes
Cursor	~/.cursor/skills/	Yes
VS Code Copilot	~/.copilot/skills/	Yes

Project Structure

python-analytics-skills/
├── .claude-plugin/
│   ├── marketplace.json    # Plugin registry metadata
│   ├── plugin.json         # Plugin configuration
│   └── CLAUDE.md           # Claude plugin guidance
├── benchmark/
│   └── pymc-modeling/      # PyMC skill benchmark harness
├── hooks/
│   ├── hooks.json          # Hook configuration
│   └── suggest-skill.sh    # Keyword-based skill suggestion
├── scripts/
│   └── validate-skills.sh  # Repository validation
├── skills/
│   ├── pymc-modeling/      # Main PyMC skill plus 13 reference docs
│   ├── prior-elicitation/  # Prior workflow references
│   ├── pymc-extras/        # pymc-extras references
│   ├── pymc-testing/       # PyMC testing skill plus testing patterns
│   ├── model-evaluation/   # LOO/model comparison references
│   └── marimo-notebook/    # Marimo references, templates, and conversion utilities
├── install.sh              # Multi-platform installer
├── package.json            # npm package metadata
└── skills.json             # Skills registry

Hooks

The plugin includes a UserPromptSubmit hook that suggests relevant skills when it detects keywords in your prompt:

• PyMC keywords: bayesian, pymc, mcmc, posterior, inference, arviz, prior, sampling, divergence, hierarchical model, gaussian process, bart, causal inference, etc.
• PyMC companion keywords: PreliZ, prior elicitation, model comparison, LOO, ELPD, pymc-extras, splines, testing PyMC, mock sampling, etc.
• PyMC testing keywords: testing PyMC, pytest, mock sampling, test fixtures, CI/CD for Bayesian models, etc.
• Marimo keywords: marimo, reactive notebook, @app.cell, mo.ui, notebook conversion, wigglystuff, etc.

Test the hook directly:

echo '{"user_prompt": "bayesian model"}' | bash hooks/suggest-skill.sh

Troubleshooting

Skill not loading:

1. Verify the skill directory exists with a valid SKILL.md
2. Run ./install.sh --validate to check structure
3. For Claude Code plugins, check claude --debug for hook/skill loading errors

Hook not firing:

1. Hooks load at session start -- restart Claude Code after changes
2. Use /hooks in Claude Code to see loaded hooks
3. Test the hook script directly with the command above

Contributing

See CONTRIBUTING.md for guidelines on adding new skills.

License

MIT License. See LICENSE for details.