Applications & Skills

Concrete things you can do with agentic tools

These are workflows that go beyond chat — they involve reading your files, writing outputs, running code, or connecting to external services. Most take 10–45 minutes to try.

★ = demonstrated in the bootcamp sessions.

Installable skills

Skills are reusable prompt-and-instruction bundles — saved recipes for tasks you do repeatedly. Instead of explaining a workflow from scratch each time, you invoke a skill and it runs.

Install the bootcamp skills

These skills are published and ready to use:

npx skills add eabeam/econ-skills

Or install one at a time:

npx skills add eabeam/econ-skills --skill econ-audit

Browse the source code: github.com/eabeam/econ-skills

Literature Review (`/lit-review`)

Structured literature search and synthesis. Give it a research question, and it searches for relevant papers, organizes them by theme, identifies gaps, and produces a formatted review memo.

Good for: starting a new project, preparing a referee report, catching up on a subfield.

Install: npx skills add eabeam/econ-skills --skill lit-review | Browse source

Econometrics Audit (`/econ-audit`)

Adversarial review of empirical claims. It reads your draft or code and flags: causal language without identification, missing robustness checks, specification concerns, and measurement issues.

Good for: self-review before submission, preparing for seminars, reviewing student work.

Install: npx skills add eabeam/econ-skills --skill econ-audit | Browse source

Data Dictionary Generator (`/data-dictionary`)

Reads a dataset’s metadata and produces a formatted data dictionary with variable names, types, labels, value ranges, and notes. Saves time on replication package documentation.

Good for: replication packages, project handoffs, codebook creation.

Install: npx skills add eabeam/econ-skills --skill data-dictionary | Browse source

Building your own skills

Pick one repeating task — grading feedback, email drafts, data cleaning, meeting prep
Do it once with AI and save the conversation
Extract the instructions into a Markdown file
Refine — add constraints, examples, edge cases as you use it

More resources: skills.sh (public skill registry) · Claude Code docs: custom commands · claudlab.in

Teaching

Build a course website from your materials

Point the tool at your syllabus, lecture notes, and schedule. It scaffolds a Quarto or Hugo site with pages for each topic, a schedule, and assignment links.

What it does: Reads your existing files, proposes a site structure, generates the pages, renders to HTML
Tool: Claude Code
Time: 1–2 hours for a working first draft

Open Claude Code in the folder with your course materials, then:

Read my syllabus.pdf and lecture-schedule.xlsx.
Build a Quarto website with:
- a home page with the course description
- a schedule page with dates and topics from the spreadsheet
- a page for each major unit with the learning objectives from the syllabus
Save everything in a new folder called course-site/.

Convert lecture notes to slides

Give it your .md or .docx notes and ask for a Quarto reveal.js or Beamer slide deck. It reads the source file, proposes a slide sequence, and writes the .qmd.

What it does: Reads your notes file, splits into slides, writes formatted output with headers and bullets
Tool: Claude Code
Time: 20–30 minutes per lecture

Read lecture-notes/week3-supply-demand.md.
Convert it to a Quarto reveal.js slide deck.
One main idea per slide. Use two-column layouts
where a graph or example would help. Keep bullet
points to 3-4 per slide max.
Save as slides/week3-supply-demand.qmd.

Grade with a rubric

Give it a rubric file and a folder of student submissions. It reads each submission, applies the rubric, and drafts feedback comments saved as individual files.

What it does: Reads rubric + submissions from your folder, writes feedback files
Tool: Claude Code
Time: 30 minutes to set up, then a few minutes per batch

Read rubric.md and all the .docx files in submissions/.
For each submission, apply the rubric. Write a feedback
file for each student in feedback/ with:
- score per criterion
- 2-3 sentences of specific feedback per criterion
- one overall comment on what to improve
Name each file feedback/[student-name].md.

Research

★ Raw data to analysis plan to code to report

The full pipeline from Session 2. Describe a dataset and research question; it proposes an analysis plan, writes Stata/R code, and drafts a summary memo — all saved as project files.

What it does: Reads your data (or its metadata), writes .do/.R scripts, generates a .md report
Tool: Claude Code
Time: 30–45 minutes for the full pipeline

Read data/cps_extract.csv. Describe what's in it:
how many observations, what variables, what's missing.

Then propose three analyses for a paper on wage gaps
by education level. Write the Stata code for the
first analysis. Save the do-file as code/wage-gaps.do
and a summary memo as output/analysis-plan.md.

Replication package assembly

Point it at your project folder. It reads your do-files or scripts, maps the dependency chain, identifies missing files, and writes a master script.

What it does: Reads all scripts in a directory, traces use/merge/include dependencies, writes master.do
Tool: Claude Code
Time: 30 minutes

Read all the .do files in this project.
Map which files depend on which — trace every use,
merge, include, and file reference.
Write a master.do that runs everything in the
correct order. Flag any files that are referenced
but don't exist.

Adversarial review of your own paper

Give it your draft as a file. It reads the full paper and flags causal language without identification, missing robustness checks, and measurement concerns.

What it does: Reads your .md/.tex/.docx draft, writes a structured referee-style report
Tool: Claude Code with /econ-audit skill
Time: 15–20 minutes

Install the skill, then run it:

npx skills add eabeam/econ-skills --skill econ-audit

/econ-audit

It will ask you to point it at your draft. Or run it manually:

Read my-paper.tex. Act as a skeptical referee.
For every causal claim, check whether the
identification strategy actually supports it.
Flag: causal language without identification,
missing robustness checks, measurement concerns,
and anything a referee would push back on.
Write the report as review/referee-report.md.

Data dictionary from a dataset

Point it at a dataset. It reads the metadata and produces a formatted codebook: variable names, types, value ranges, labels, notes.

What it does: Reads dataset headers/metadata, writes a formatted .md data dictionary
Tool: Claude Code with /data-dictionary skill
Time: 10–15 minutes

Install and run:

npx skills add eabeam/econ-skills --skill data-dictionary

/data-dictionary

Or manually:

Read the headers and first 20 rows of data/survey.csv.
Produce a data dictionary with: variable name, type,
range or categories, number of missing values, and
any notes about coding or units. Save as docs/codebook.md.

Literature review with structured output

Give it a research question. It searches, organizes papers by theme, identifies gaps, and writes a formatted review memo saved to your project folder.

What it does: Searches, reads abstracts, writes a structured .md review memo
Tool: Claude Code with /lit-review skill
Time: 20–30 minutes

Install and run:

npx skills add eabeam/econ-skills --skill lit-review

/lit-review

Or manually:

I'm starting a project on the effect of cash transfers
on educational outcomes in sub-Saharan Africa.
Search for recent papers. Organize by methodology
(RCT vs quasi-experimental). For each paper, note:
authors, year, country, method, main finding.
Identify gaps. Save as lit-review/cash-transfers-education.md.

Writing & Communication

★ Create a writing voice file

Give the tool access to files with your real writing — saved emails, letters of recommendation, feedback comments. It reads them, describes your patterns, and saves a reusable voice file.

What it does: Reads your writing samples from disk, writes MY-VOICE.md
Tool: Claude Code
Time: 15 minutes to create, then reuse forever

Save 5–10 of your sent emails as .txt or .md files in a folder, then:

Read all the files in writing-samples/.
These are emails and letters I've written.
Describe my writing style: tone, sentence length,
how I open and close messages, level of formality,
what I avoid, any distinctive patterns.
Save the description as MY-VOICE.md.

Then tell Claude Code to use it going forward:

Add to your memory: when I ask you to draft anything,
load MY-VOICE.md first and match my style.

★ Draft emails using your voice file

Point it at an email thread (via Gmail integration or a saved file) and tell it the gist. It reads the thread, loads your voice file, and drafts a reply that sounds like you.

What it does: Reads email thread + voice file, writes a draft
Tool: Claude Code with Gmail integration
Time: 2 minutes per email

Read my voice file. Read the most recent email
from [sender] about [topic].
Draft a reply. The gist: [your one-sentence summary
of what you want to say].

Revision memo for a journal submission

Give it your paper file and the referee reports. It reads both and drafts a point-by-point response memo.

What it does: Reads your draft + referee reports, writes a structured response .md file
Tool: Claude Code
Time: 30–45 minutes for a first draft

Read paper/draft-v2.tex and reviews/referee1.pdf
and reviews/referee2.pdf.
Draft a point-by-point response memo. For each
referee comment:
- quote the comment
- describe what we changed (or why we didn't)
- reference the specific section/page
Save as paper/revision-memo.md.

Workflow & Organization

Inbox triage

Connects to your Gmail, reads recent messages, and categorizes them by action needed: reply, review, meeting, delegate, archive.

What it does: Reads inbox via Gmail integration, writes a categorized summary
Tool: Claude Code with Gmail integration
Time: 30 minutes to set up, 5 minutes each morning

Read my 20 most recent unread emails.
Categorize each one:
- 🔴 Needs reply (from a person, expects a response)
- 📋 Needs review (requires reading but not a reply)
- 📅 Meeting-related (scheduling, agenda, follow-up)
- 📦 Can archive (newsletters, notifications, FYI)
List the reply-needed ones first with a one-line summary.

Calendar automation from emails

Reads emails with dates and creates calendar events automatically.

What it does: Reads email via Gmail, extracts dates/times, creates Google Calendar events
Tool: Claude Code with Gmail + Google Calendar integrations
Time: 5 minutes per batch

Search my Gmail for emails from [hockey league / school]
in the last 7 days. Extract any dates, times, and
locations for upcoming events. Add each one to my
Google Calendar with the event name, time, and location.

Project handoff notes

At the end of a work session, it summarizes what was done, what’s open, and what to do next. Saves as a file so you can pick up later without re-reading everything.

What it does: Reads session context, writes a structured handoff .md file
Tool: Claude Code
Time: 2 minutes

Summarize this session:
- what we worked on
- what decisions were made
- what's still open
- what I should do next
Save as notes/handoff-2026-04-08.md.

What to try first

Fastest payoff: Create a voice file (15 minutes, reuse forever)
Most impressive: Run the data → code → report pipeline (30 minutes)
Most practical: Inbox triage with Gmail integration (30 minutes to set up, then daily)