Workshop schedule

§ i

Mon 1 Jun

Foundations & first real workflow

Deliverable
Working Claude Code env + your own global CLAUDE.md + a cleaned, aggregated dataset with assertion checks.

• ◦Landscape: agentic vs. chat; harness vs. model
• ◦Setup, first conversation, permission model, slash commands
• ◦Install R or Python and the workshop packages, using Claude Code itself
• ◦What plays well with agentic tools (formats, languages)
• ◦Plan / Execute / Clear; lightweight assertions
• ◦Data profiling → cleaning → aggregation on workshop data
• ◦Interview trick: generate your own global CLAUDE.md

§ ii

Tue 2 Jun

From data to compiled output

Deliverable
Regression script with assertions + publication-quality table + figure + Markdown PDF.

• ◦Three prompting principles + three power prompts
• ◦Live regression demo: fixest, FE, clustered SE, multi-spec
• ◦Tables: encode conventions, never hand-finish
• ◦Figures: ggplot2, coefficient plots, iterative refinement
• ◦Compile-fix loop: Markdown via pandoc; LaTeX as parallel track
• ◦Trust-but-verify exercise: prepared script with subtle bugs

§ iii

Wed 3 Jun

Customizing & extending Claude Code

Deliverable
Multi-level CLAUDE.md + memory + 2 custom commands + Superpowers installed + 1 custom skill.

• ◦The customization stack mental model
• ◦Multi-level CLAUDE.md (global / project / subdirectory)
• ◦Memory system: instructions vs. facts
• ◦Custom slash commands: encoding repeated prompts
• ◦Install Superpowers + Crawfurd's paper-review skill
• ◦Build a custom skill for your own workflow
• ◦Git worktrees for parallel robustness specs

§ iv

Thu 4 Jun

Applied research & capstone

Deliverable
Working API script + meaningful progress on your own research project.

• ◦API demonstrations: FRED in R, OpenAI text classification
• ◦Drafting prose with Claude (collaboration, not ghostwriting)
• ◦Applied capstone: 90 minutes of focused work on your own research
• ◦Participant share-out, 30-day plan, closing remarks