Resources

These are the outside tools, templates, and readings behind the course. It is the same vetted list the module outlines were built from, grouped by the module where each item is most relevant. Each item gets one line on when an accounting PhD would actually reach for it. Nothing here is required reading. These are pointers for after the course ends, when you’re building your own pipeline rather than following the lab.

One item is flagged rather than recommended. It is listed because it exists and people are talking about it, not because it has been vetted to the standard of everything else on this page. Agents that helped build this course were allowed to study any of these workflow templates for patterns, but never to copy prose from them. The same no-verbatim rule governs this course’s use of its own primary source material (see About).

Module 1 — setup and mental models

  • hanlulong/AI-research-setup — step-by-step Claude Code + Codex setup for economics research on macOS and Windows. Reach for this before a build module if the course’s own Setup page leaves a platform-specific gap, especially on Windows.

  • Korinek, AI Agents for Economic Research (NBER w34202, 2025) and genaiforecon.org — the source for this course’s agent-framing lecture. Reach for the site for its semi-annual tool updates, since a course syllabus goes out of date faster than a maintained site does.

  • pedrohcgs/claude-code-my-workflow — Pedro Sant’Anna’s Emory Econ 730 template (contractor mode, quality gates, session logging). Reach for this when you’re setting up your own lab’s standing Claude Code conventions and want a second worked example beyond this course’s own CLAUDE.md.

  • Dell’Acqua, McFowland, Mollick, Lifshitz, Kellogg, Rajendran, Krayer, Candelon and Lakhani, “Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality,” Organization Science (2026), doi:10.1287/orsc.2025.21838 — the field experiment with Boston Consulting Group consultants behind Module 1’s point that AI help can improve some tasks and quietly worsen others of similar difficulty. Reach for this when you are deciding how closely to check an agent’s work on a given step.

Modules 1/5 — ready-to-fork research workflow templates

These show what a mature setup looks like once you’re past the basics. Borrow CLAUDE.md and quality-gate patterns from them rather than adopting one wholesale.

  • nemmerel/claude-workflow — an economist-oriented Sant’Anna fork with econometrics-reviewer, replication-auditor, and referee-response agents spanning the full research lifecycle. Reach for this when you want a template for a specific agent role (a replication auditor, say) rather than building one from scratch.
  • chesun/claude-research-workflow — an empirical social science template built around an explicit “epistemic stack” (no-assumptions, primary-source-first, derive-don’t-guess, adversarial-default). Reach for this alongside the Verification page. Its epistemic-stack framing pairs directly with this course’s verification-debt lecture.
  • eweisbrod/example-project — Eric Weisbrod’s reproducible-project template for accounting research (R, Python, SAS, and Stata), including a dedicated page on meeting JAR’s data-sharing policy: replication code, SAS-log-style execution logs, and exported sample identifiers with file hashes. Reach for this when a paper is headed to JAR, or any journal with a data policy, and you want to see Module 5’s paper-trail discipline mapped onto an actual submission requirement.
  • benediktheid/clo-author — a lit-review-to-submission pipeline with adversarial QA built in. Reach for this if you’re scoping an end-to-end drafting pipeline for a full paper, not just one section.
  • izhiwen/AiEconLab — a multi-role “virtual research team” (PI/RA/referee personas). Watch, unvetted: this is new and unproven (near-zero stars at time of writing). It is listed so you know it exists, not as a course-vetted recommendation.

Module 3 — EDGAR / text-as-data

  • stefanoamorelli/sec-edgar-mcp — an SEC EDGAR MCP server. Reach for this after you’ve built Module 3’s scrape by hand and want to compare building it yourself against adopting an existing MCP. The comparison shows you the tradeoff between the two; the MCP is not only a shortcut.

Module 4 — Stata bridge and data MCPs

  • hanlulong/stata-mcp — a VS Code/Cursor Stata MCP extension with real-time output, a data viewer, and graphs. Reach for this if you want the Parquet/DuckDB/Stata handoff from Module 4 to happen inside an IDE rather than the terminal.
  • tmonk/mcp-stata — a lightweight Stata MCP with a skills catalog for replication, robustness, specification comparison, and publication QA. Reach for this as the closest existing analogue to this course’s own wrds filter skill and its iron-law checklist.
  • SepineTam/stata-mcp — a third Stata MCP option. Compare it against the two above before you commit to one for a lab’s standing toolchain.
  • Scott Cunningham’s Claude Code series on Causal Inference — a running set of posts on using Claude Code for applied econometrics, including Claude Code 31, which runs one difference-in-differences estimator through six packages in R, Python and Stata and gets different numbers. Reach for this when you set up the two-language check from Module 4 and want to see how far implementation defaults can move a result.
  • Public-data MCP zoo (not taught in the course, but they show the pattern generalizes): fred-mcp-server, worldbank/data360-mcp, US Census MCP, financial-datasets/mcp-server. Reach for any of these when a project needs macro, trade, or census-style public data through the same MCP pattern Module 4 teaches for WRDS, without a WRDS account behind it.

Module 5 — writing and review

  • hanlulong/econ-writing-skill — compare it against this course’s own barrios-voice and econ-humanizer skills to see a different approach to the same problem.
  • claesbackman/AI-research-feedback — structured AI feedback on drafts. Reach for this as a second input alongside Module 5’s editor-not-rewriter workflow when you want a differently structured pass at the same section.
  • scunning1975/MixtapeTools — Scott Cunningham’s collection of Claude Code personas and skills, including the Referee 2 persona behind Module 5’s clean-room referee. Reach for this when you want a worked example of an adversarial review prompt to adapt to your own field’s norms.
  • aspi6246/Claude-Code-Presentation — Alessandro Spina’s finance brownbag on Claude Code, with an “Editor” persona that audits a paper in passes and has a mode for checking a response letter against the manuscript. Reach for this to see the idea behind Module 5’s response-letter audit. No license: the repository grants no reuse rights, so read it for ideas and credit them, but don’t copy its text or slides.
  • Future-House/paper-qa and papersgpt-for-zotero — lit-review tooling. Reach for either when a literature synthesis task has outgrown a single chat conversation and needs to query a real library of PDFs.
  • craigsc/cmux — “tmux for Claude Code”: one command per parallel agent, each in its own git worktree. Reach for this when a revision forks, for example robustness variants explored side by side, or a pipeline rebuild running while a second session works on the prose. See the Tools page for the workflow and the verification caveat.
  • manaflow-ai/cmux — same name, different tool: a Ghostty-based macOS terminal with vertical tabs and notifications, built for watching several coding agents at once. Reach for this if you run parallel sessions daily and want one screen to watch them all, not just the workflow.

Other people’s lists

For when this page goes out of date: awesome-ai-for-economists, meleantonio/awesome-econ-ai-stuff, Will Rinehart’s AEI AI-tools guide, and Cowen & Tabarrok’s “How to Learn and Teach Economics with LLMs.” Other people maintain these on their own schedule, so check them now and then, independent of this course.

Everything on this page was vetted when this course was built. Unvetted items are labeled as such rather than included without comment. See About for the attribution policy governing how these resources were used while building the course materials.