Beyond Prompting: Using Agent Skills in Data Science

Agent skills are reusable packages of instructions (centered on a SKILL.md file) that let AI coding tools like Claude Code or Codex handle recurring workflows more reliably than one-off prompts. The author demonstrates this by packaging an 8-year weekly data visualization habit into two skills: one that analyzes datasets, surfaces insights, and generates interactive charts, and another that publishes them to a personal website. Building the skill required iterative refinement — sharing personal style guidelines, researching external best practices, and testing across 15+ datasets. Skills pair well with MCP for tool access, and are most valuable for tasks that recur, follow semi-structured processes, and depend on domain knowledge.