AI Tools for Academic Work

Here are some AI tools that have been very helpful to me as an academic, which may be helpful to others! Beyond the now-widespread uses on coding and writing, these tools can help on things specific to academic administration work (e.g. filing for reimbursements following the specifics of your institutional policies, or updating your CV from your own records). If you find bugs or have suggestions on how to improve these tools, don't hesitate to reach out!

The main thing I found myself needing to build into these tools is rigor. What that means in practice looks different across tasks:

  • Citing literature — the agent may not cite a paper it hasn't actually read; every reference is verified to exist, or flagged for you to check; everything it reads goes into a running log, so future sessions build on the same body of knowledge.
  • Coding — scripts are read end-to-end and audited for internal consistency; Stata syntax is verified against the official manuals before writing; and for data work in secure-enclave environments, queries are written to avoid wasting scarce computing resources.
  • Administrative work — the agent internalizes your institution's actual policy document and checks every claim against it, rule by rule, deadline by deadline.
  • And when it's unsure, it asks — ambiguity is flagged back to the user instead of resolved independently.

Note: None of these tools are designed to be token-efficient; rather, these workflows are designed to be conservative and meticulous, which in my experience is exactly what cuts down the frustrating iteration loops with AI tools.

View on GitHub Quickstart

The Five Workflows

  • Administrative Reimbursements, receipts, and updating CV/bio.
    What's inside

    /ingest-policy reads your institution's reimbursement-policy PDF and generates a rule-by-rule compliance rulebook (portable across institutions). /update-document updates a CV/bio based on activity on your computer — e.g. a referee report for a new journal prompts you to add the journal under your Refereeing Service section. Specialist agents classify receipts, fill the claim form, review it against policy, and triage deadlines.

  • Academic Writing Construct a body of knowledge based on the relevant literature, appropriately cite relevant authors and find relevant papers (via the paired lit-review skill — see Resources), and position papers within the literature.
    What's inside

    Hard rules for rigor: no-fabrication (every claim verified against its source, with a mechanical self-check), lit-review-protocol (every paper read is logged before it’s cited), and replication-protocol (reproduce published results before extending). Plus a notation-registry convention for theory papers.

  • Coding with SNDS Analysis in the French SNDS secure health-data enclave (SAS/Oracle + R).
    What's inside

    SAS/SQL conventions (partition keys, composite joins, safe-replace), and DUA-compliant confidentiality: statistical-disclosure-control and export-compliance rules, a sas-reviewer agent, and a /sds-doc lookup against the authoritative SNDS documentation.

  • Coding with Stata General empirical analysis in Stata (+ R for figures).
    What's inside

    Do-file conventions, a stata-reviewer agent, and /stata-syntax — which verifies command syntax against the manuals before writing, because a syntax error wastes a run.

  • Teaching Pedagogy-oriented tools to keep track of project supervision, problem set design, and reading lists.
    What's inside

    /supervise-project (flags the advisor to check in with the student, and tracks feedback given/content reviewed), /tf-builder (teaching-fellow problem sets, with solutions in a separate file), and /reading-list (curated, citations verified or flagged).

How It Works

Workflows are self-contained .claude/ configs that bundle their instructions, rules, hooks, and — where their skills need them — specialist agents. Two cross-cutting rigor skills back them: /rigor (never fabricate parameters; cite or stop — auto-loaded every session, in every workflow) and /empirical-coding-discipline (audit every step; verify units and identifiers — used by the coding and writing workflows). A shared working discipline rides in every workflow: plan first, verify after, 80/90/95 quality gates, session logging. For the general academic layer these pair with — lecture production, literature review, generic paper review — see the resources below.

# 1. make the shared skills available (once per machine)
cp -R skills/* ~/.claude/skills/

# 2. drop a workflow into a project
cp -R workflows/academic-writing-workflow/.claude /path/to/project/.claude
#    then open Claude Code and fill the [YOUR …] placeholders

Nothing assumes git or Quarto; the templates are LaTeX/Beamer + R/SAS/Stata oriented. Full steps are in the repo's INSTALL.md.

Download

The repository is available on GitHub —

github.com/acquat/claude-academic-workflows

Everything here runs on Anthropic's Claude Code. Codex version coming soon!

Found it useful, or improved something? I'd love to hear about it!

Other Helpful Resources

Many great AI tools are being built by academics, for academics — these are the ones I use and recommend:

  • claude-code-my-workflow — Pedro Sant'Anna's comprehensive academic foundation (slides, lit review, paper review, data analysis). My own setup grew out of this kit, and these workflows are designed to install alongside it. In particular, download his /lit-review skill to use with my academic-writing workflow: his skill searches and synthesizes the literature, while my logging rule makes sure every paper it reads is recorded — with the pages read and the load-bearing claims — before it can be cited. They pair naturally.
  • stata-skill — Dylan Moore's Stata reference skill; a complementary approach to the same "Claude can't run Stata" problem.
  • strategic-revision — a super useful revision planner for revise-and-resubmits, by GitHub user jusi-aalto.
  • "Automatic" logging in Claude Code — Michael Ewens' session-logging hook.