How I currently use Claude Code. This is a living document that evolves with my workflow.

Environment#

• Terminal: cmux
• Dictation: Wispr Flow for voice input. I have a note about interpreting phonetic typos in CLAUDE.md.
• Config managed via dotfiles repo (~/.claude/ symlinked into ~/git/dotfiles/claude/)

I use cmux. I used to use bare Ghostty splits (⌘+D / ⌘+⇧+D), and before that tmux (see tmux + Neovim setup). My ranking: cmux > bare Ghostty > Ghostty + tmux. cmux is built on top of Ghostty, so I still get its speed and native macOS integration. Bare Ghostty also had an annoying tab-naming issue: since each tab is a project, I want stable names, but Ghostty changes the tab title based on whatever is running in the focused split. cmux gives me a sidebar for project-based organization (similar to the Codex app) with stable naming, while keeping things lightweight. Its split and multi-tab logic feels more natural: tabs are created within a split rather than splits being tied to a single tab, which better matches how CC workflows branch out.

My default layout is two panes side by side — Claude Code on the left, Neovim on the right. I split further as needed but try to return to this two-pane setup. Neovim is the right fit here for the same reason as Ghostty: snappy and versatile. You can quickly move around, explore files, and make edits while CC works in the other pane. I sometimes run watch scripts (e.g., auto-compile, file sync) in a tiny extra pane.

Usually I have several tabs open (one for each project) with each tab having 2+ panes.

CLAUDE.md setup#

• Global (~/.claude/CLAUDE.md): Python/uv preferences, wiki/blog paths, project tracking conventions, academic writing style
• Project (CLAUDE.md): repo-specific instructions, commands, code style

Custom skills#

Most of my research-oriented skills are packaged as Claude Scholar (yy/claude-scholar):

• /verify-math — step-by-step verification of mathematical derivations using SymPy
• /doi-bibtex — grab BibTeX entries from DOIs
• /arxiv-metadata — fetch structured metadata for arXiv papers
• /openalex — query 240M+ scholarly works via the OpenAlex API
• /check-refs — verify every citation in a LaTeX paper against academic databases
• /latex-cleanup — catch common LaTeX issues: orphan lines, spacing, typography, draft artifacts
• /presubmit-checks — pre-submission checklist running multiple checks in parallel
• /arxiv-prep — prepare an arXiv submission package

Other personal skills (not in claude-scholar) include:

• garden — Wiki gardening—pick a page and improve it
• img2s3 — upload an image to S3

Plus the Anthropic document-skills plugin for PDF, PPTX, DOCX, etc.

Custom agents#

• file-paper — file an academic paper to my wiki. It fills in metadata and then searches my wiki to identify & link relevant papers and concepts
• wikify — turn anything into a wiki page. Calls file-paper when necessary, otherwise follows similar steps to create links to relevant pages
• update-paper-status — update a paper’s publication status in CV bib file
• codebase-analyzer, code-smell-refactor, security-auditor — code quality

Custom commands#

• sync-repo — add, commit, pull –rebase, push
• sync-all — auto-commit and push all git repos before switching computers
• release — bump version, tag, push, create GitHub release, publish
• uv-init — bootstrap Python/uv/direnv environment
• new-research-project — scaffold from my project template
• launch-checklist — launch an OmniFocus template from any project
• wikify — routes to the right agent (file-paper for papers, wikify for everything else)