Resources

Claude Code is Anthropic’s coding agent, and this is the research workflow I have built on top of it. Fourteen skills cover most of the empirical cycle: finding and reading papers, running a literature review, auditing a .bib file against canonical metadata, drafting a preregistration, checking a manuscript before submission, writing an R&R response, assembling a replication archive, compiling LaTeX, and iterating on TikZ figures until they look right. bibcheck catches the wrong years, mis-cited authors, and hallucinated references that survive a spell-check. tikz-iterate hands each rendered figure to a critic agent that looks at the picture rather than the code.

It also ships my global CLAUDE.md, the standing instructions loaded into every session, and a Quarto reveal.js slide system whose quality gates catch an overflowing slide before you find out at the podium. Quarto does four things that Beamer and PowerPoint do not do together: real animation, LaTeX from macros you define, R and Python that execute at render, and citations straight from a .bib file.

A closing section, Things you may not know, covers the parts that changed how I work day to day: the agent view for running many named sessions side by side, hold-to-talk voice input, steering a session from your phone while you are away, Overleaf projects as plain local folders, and the connectors worth setting up.

Everything there, including the two example decks, was generated with Claude Code as a proof of concept, so verify citations, numbers, and claims before relying on any of it.