Resources from Ben Lira's one-hour Claude Code demo to the Wharton marketing department (2026-04-16). Share freely.
.
├── install.sh # One-liner installer (see below)
├── CLAUDE.md # Context for future Claude sessions in this repo (the demo prep)
├── TEMPLATE_CLAUDE.md # Starter template for YOUR own global ~/.claude/CLAUDE.md
├── demo_script.md # Full runbook: what was demoed, exact prompts used, Q&A answers
├── slides.qmd # Quarto source for the slide deck
├── slides.html # Rendered slides (open in browser)
└── skills/ # 16 slash-command skills
├── archive-raw-data/
├── aspredicted/
├── data-analysis/
├── gws/
├── latex-gdoc-roundtrip/
├── latex-workflow/
├── log-off/
├── notify-me/
├── opportunity-hunter/
├── pipeline-audit/
├── plate-check/
├── qualtrics-survey/
├── quarto-version-toggle/
├── render-survey/
├── research-writing/
└── wbl-form-filler/
One-liner (installs all skills into ~/.claude/skills/):
curl -fsSL https://raw.githubusercontent.com/lirabenjamin/claude-code-for-researchers/main/install.sh | bashInstall only specific skills:
curl -fsSL https://raw.githubusercontent.com/lirabenjamin/claude-code-for-researchers/main/install.sh | bash -s -- pipeline-audit data-analysisRe-install and overwrite existing:
curl -fsSL https://raw.githubusercontent.com/lirabenjamin/claude-code-for-researchers/main/install.sh | bash -s -- --forceThe script clones the repo to a temp dir, copies skill folders to ~/.claude/skills/, skips any that already exist (unless --force), and prints a list of skills that need further configuration.
Manual alternative — clone the repo and copy what you want:
cp -r skills/pipeline-audit ~/.claude/skills/After installing, restart Claude Code (or start a fresh session). Skills appear as slash commands, e.g. /pipeline-audit. Read the SKILL.md first — several skills use placeholder variables like {{WORKSPACE}} or {{PHONE}} that you need to replace. grep -l '{{' ~/.claude/skills/*/SKILL.md will find them.
| Skill | What to replace |
|---|---|
archive-raw-data |
.env needs ZENODO_TOKEN (from https://zenodo.org/account/settings/applications/tokens/) and MONGODB_URI. Requires mongoexport installed. |
plate-check |
{{WORKSPACE}} (path to your comms/todo/current_state directory), {{PHONE}} (iMessage number or delete the notify step) |
log-off |
{{WORKSPACE}}, {{PHONE}} (same as plate-check) |
opportunity-hunter |
{{REPORTS_DIR}} (where opportunity reports get saved) |
wbl-form-filler |
All defaults in the "Known Defaults" block (your name, email, faculty contact, IRB, budget codes). Wharton-specific — adapt for your own lab's forms. |
qualtrics-survey |
Needs QUALTRICS_API_TOKEN and QUALTRICS_DATA_CENTER in a .env file |
render-survey |
Needs Render + MongoDB Atlas accounts + OPENAI_API_KEY if using AI tasks |
notify-me |
Uses macOS Messages + your phone number |
gws |
Requires gws CLI installed for Google Workspace integration |
The other skills (aspredicted, data-analysis, latex-gdoc-roundtrip, latex-workflow, pipeline-audit, quarto-version-toggle, research-writing) work out of the box.
TEMPLATE_CLAUDE.md is a starter template for your global ~/.claude/CLAUDE.md. Copy it over and fill in the placeholders:
cp TEMPLATE_CLAUDE.md ~/.claude/CLAUDE.md
# then open and fill in your detailsThis is the single highest-leverage thing you can do. Once Claude knows your role, active projects, technical preferences, and failure modes, every session starts with context — you stop repeating yourself.
-
demo_script.md— full runbook with time-budgeted sections, exact prompts, fallback plans, and scripted answers to the two audience questions (AsCollected data provenance; hallucination checks in analysis code). -
slides.html— 10 concept slides + 4 section-transition slides. Rendered fromslides.qmd. To re-render after editing:quarto render slides.qmd
File issues on this repo or email Ben. If a skill breaks or a placeholder is missing, flag it.
These skills are shared freely. Adapt, fork, improve, re-share. No warranty — you are responsible for checking what Claude produces. See demo_script.md Q&A section for Ben's specific answer on how to verify AI-generated analysis code.