A set of skills to call your agent Bayes. Thomas Bayes.
Agent Skills for Bayesian modeling, causal inference, and probabilistic thinking. Compatible with Claude Code, Kimi Code, Cursor, Gemini CLI, and any agent that supports the Agent Skills spec.
| Skill | Description |
|---|---|
| bayesian-workflow | Full Bayesian modeling workflow with PyMC and ArviZ. Full breakdown. |
| causal-inference | Production-grade Bayesian causal inference with PyMC, CausalPy, and DoWhy. Full breakdown. |
| amortized-workflow | Amortized Bayesian workflow with BayesFlow for simulation-based inference. |
More skills coming soon. Issues and PRs are welcome!
git clone https://github.com/Learning-Bayesian-Statistics/baygent-skills.git /tmp/baygent-skills
mkdir -p ~/.claude/skills
cp -r /tmp/baygent-skills/bayesian-workflow ~/.claude/skills/
cp -r /tmp/baygent-skills/causal-inference ~/.claude/skills/
cp -r /tmp/baygent-skills/amortized-workflow ~/.claude/skills/ # BayesFlow / SBIClone the repo and copy the skill folders you need into your agent's skills location:
git clone https://github.com/Learning-Bayesian-Statistics/baygent-skills.git /tmp/baygent-skills
cp -r /tmp/baygent-skills/bayesian-workflow/ ~/.config/agents/skills/bayesian-workflow/
cp -r /tmp/baygent-skills/causal-inference/ ~/.config/agents/skills/causal-inference/
cp -r /tmp/baygent-skills/amortized-workflow/ ~/.config/agents/skills/amortized-workflow/Note: causal-inference depends on bayesian-workflow — install both if you want causal inference.
These skills are opinionated and workflow-first. They don't just teach an agent what PyMC functions exist — they enforce a specific sequence of steps (prior predictive checks, diagnostics, calibration, reporting) and guardrails (94% HDI, reproducible seeds, save-to-disk) that produce reliable analyses.
Each skill is focused and lean. Rather than one monolithic skill that covers everything, we build specialized skills that do one thing well:
- bayesian-workflow covers the fundamentals that every Bayesian analysis needs.
- causal-inference handles causal design, identification, and refutation — delegating the modeling to bayesian-workflow.
- amortized-workflow covers simulation-based inference with BayesFlow — end-to-end architecture selection, training, simulation-based diagnostics, and real data application.
Created by Alexandre Andorra, host of Learning Bayesian Statistics.
MIT - see LICENSE.