Maintainers: Pedro H. C. Sant'Anna (Emory University, Economics)
Compatibility:claude-code
A ready-to-fork Claude Code template for academics using LaTeX/Beamer + R. 61 skills + 18 agents. Calibrated peer-review pipeline against AER, QJE, JPE, Econometrica, ReStud. Adopted by 15+ research groups. Refresh 2026-09-29 (upstream at 2026-09-27, previously catalogued at 2026-04-27 with 30 skills + 14 agents): added 31 skills (replication packaging, environment capture, disclosure and submission checks, grant/DMP/power-analysis design aids, teaching tools, audit and verification primitives such as vaccinate, differential-audit, verify-artifact, oracle-review) and 4 agents (humanize-auditor, promote-memory-council, r-package-reviewer, sim-reviewer); 43 existing skill texts changed upstream and were re-copied (notably /commit no longer merges, /deep-audit became a general adversarial audit); nothing was removed.
Stress-tests a finding against the analysis forks not taken (measures, sample filters, controls, clustering, weighting, functional form) by running and reporting the specification grid.
Root-causes a failing or wrong empirical result with a reproduce, minimise, hypothesise, instrument, fix loop tuned for R, Stata and Python research code.
Scaffolds and runs a reproducible Monte Carlo simulation study in R (DGP, estimator grid, seeded replications), reporting bias, RMSE, coverage and size/power with Monte Carlo standard errors.
Before and after changing anything shared (a function, signature, schema, config default or constant), finds every consumer and runs it to catch silent downstream breakage.
Research-brief and claim-record discipline for AI-assisted research: write the brief before a long run, and back every reported result with a claim record.
Adversarial audit of a theory, proof, paper, codebase or set of claims: independent skeptics must return concrete defects, a separate judge adjudicates each, confirmed defects are fixed and re-verified.
Compares two implementations of the same thing (a port, reimplementation, replication package or refactor) with frozen inputs and a full output inventory so agreement is meaningful.
Pre-screens tables, figures and logs built on restricted data for statistical-disclosure problems (small cells, suppression gaps, dominance, re-identification risk) before release.
Qualifies a checker before trusting it: seeds known defects into a copy of a real artifact plus a clean control and reports recall and false-positive rate.
End-to-end Stata pipeline: numbered .do files run through the stata-mcp server, logged outputs, and publication-ready esttab tables and exported figures.
Computes power, required sample size and minimum detectable effect for a study design (clustered RCTs, multiple arms, simulation-based) and writes a registry-ready power section.
Scaffolds a research grant proposal (NSF, NIH, ERC or foundation) by composing the interview spec, data-management-plan and environment-capture skills.
Builds or restructures a course syllabus: description and prerequisites, week-by-week schedule, measurable learning objectives, assessment scheme with rubric, and standard policies.
Generates a co-author handoff brief: git delta since the last brief, state of manuscript, analysis and slides, open questions, and how to reproduce locally.
Runs the quality and consistency gates, then commits with a subject that states what is now true; pushes and opens a PR only on request and never merges without an explicit instruction.
Keeps the project to-do list and memory in GitHub issues: files findings without duplicates, lists issues relevant to the current work, and closes them with a record of the fix.
Reviews candidate learnings in Claude Code auto memory through a five-critic council and promotes majority-approved entries to the committed MEMORY.md.
Five-critic council (generality, staleness, redundancy, evidence, format) that votes on promoting candidate learnings from local memory to the committed MEMORY.md; invoked by /promote-memory.
Triages academic email and calendar into a prioritized digest plus a referee-obligations tracker (referee requests, R&R correspondence, co-author threads, invitations).
Scaffolds a new skill in the repo conventions: interviews for purpose, trigger phrases and tool needs, then writes a SKILL.md that passes the integrity gates.
Assembles a submission-ready replication package to the AEA Data and Code Availability Standard: README, dataset manifest, computational requirements, table and figure map.
Runs an external frontier-model referee from a different vendor (via the Oracle CLI) on a paper, proof, estimator or replication package, then triages each finding as confirmed, refuted or downgraded.
R package source reviewer for CRAN readiness: DESCRIPTION and dependency hygiene, NAMESPACE, roxygen completeness, testthat coverage and CRAN-policy red flags.
Monte Carlo simulation reviewer: assumption regime, DGP and estimand alignment, replication budget and Monte Carlo SE, coverage against the truth, and parallel-seed discipline.
Treats every incoming review finding (referee, AI reviewer, linter, second model) as a candidate, checks it against the source, and applies only verified fixes.
Turns a research paper into teaching materials: lecture outline, the key results with intuition, a slide skeleton for /create-lecture, discussion questions and a problem-set brief.
Generates the submission-time disclosure block: AI-use statement matched to the target journal policy, CRediT roles, conflict-of-interest and data-availability statements.