Skills Cookbook
Lab guide for adopting agentic work and coding assistants.
Tutorials
Setup
Get started with AI coding tools — install OpenCode, connect to OpenRouter, and run your first Kimi K2.5 prompt.
Penguins Analysis
Data analysis with AI assistants — explore a real dataset, write code with suggestions, and produce publication-ready plots.
Journal Club Slides
Create presentations from papers — feed a PDF to an agent and receive a structured Reveal.js slide deck ready to present.
Single Cell Portal
Build an interactive scRNA-seq viewer — scaffold a Shiny/web app that lets users explore dimensionality-reduction plots.
Context Management
Understand context windows and costs — learn how context rot develops and how to use /compact to keep runs cheap and accurate.
Figure Legend Generator
Workshop challenge — build an agent that writes publication-ready figure legends. Best generator wins.
APM
Dependency manager for AI context — use the Agent Package Manager to declare, version, and inject context modules automatically.
Second Brain
Build a personal research wiki from your own papers — drop PDFs in raw/, have the agent compile summaries, concept articles and backlinks, then query across it.
Second Brain — Stretch Goals
Once your wiki works — run a linting pass, add the claude-obsidian plugin, render matplotlib figures from frontmatter, turn answers into Marp decks, vibe-code a search engine.
Pharmacogenomics with ClawBio
Install ClawBio, run a pharmacogenomics report on demo (or your own) genetic data, and have your agent translate CYP2D6 metabolizer status into CPIC dose-adjustment recommendations.
Prerequisites
- GitHub Education — Free Copilot Access — Apply for GitHub Education benefits and activate Copilot Pro with Claude
- Get OpenCode — Download and Connect to Copilot — Install OpenCode, sign in with GitHub, and pick Claude Opus 4.5
Extra Resources
- Anthropic’s Guide to Building Effective Agents
- claude-mem: Memory System
- Claude Code: A Highly Agentic Coding Assistant (DeepLearning.AI)
Motivation
“Nothing we do as we know it will be the same in three months”
“Code is over”
“We should be burning through a data centre’s worth of credits” — Nathan