GPT Sherpa
Practical patterns for working with LLMs.
Practical patterns for working with LLMs
A working data scientist’s notebook of patterns that hold up in production: how to set up Claude Code, how to write CLAUDE.md control documents that make an LLM behave, how to author Quarto documents with LLM help, and a small library of templates you can copy.
Not generic LLM hype, not vendor docs — patterns proven on real projects, generalized for anyone to reuse.
Three places to start
New to Claude Code?
The setup guide walks through installing the CLI, configuring permissions, and adding your first custom skill.
Already running Claude Code?
Make it work better. A CLAUDE.md control document is how you teach an LLM your project. Start from the template, refine as the project grows.
What’s here
- Getting started — what LLM-assisted work looks like; a quick GitHub Copilot primer.
- Claude Code — setup and an opinionated architecture-review pattern for organizing projects with Claude.
- Quarto — authoring skill, an LLM cheatsheet, and the ontology-website pattern.
- Skills & patterns — drop-in skill files for AskSage multi-model review, data profiling, PHI scanning, knowledge vaults, txtarchive, Rhino + Shiny, and tidymodels.
- Templates — project and global
CLAUDE.mdstarters, a permissionssettings.local.json, an ontology project scaffold, and a Quarto research-paper starter pack.
Source: llmcheatsheets on GitHub (MIT licensed).