arXiv Assistantan Omarchy plugin

Configuration

Settings live on the widget's entry in ~/.config/omarchy/shell.json and apply as soon as the file is saved. The quickest way to change one is omarchy bar set:

omarchy bar set io.github.pierrecagne.arxiv categories "hep-th, gr-qc"
omarchy bar set io.github.pierrecagne.arxiv viewerPort 8795 --json

An entry in shell.json looks like this:

{
  "id": "io.github.pierrecagne.arxiv",
  "categories": "math.CT, cs.LO, cs.PL",
  "reader": "a working mathematician in categorical logic, categorical semantics and type theory",
  "reportStyle": "Theory",
  "engine": "Claude Code",
  "sandbox": true,
  "extraReadDirs": "~/research/notes"
}

All settings

KeyDefaultMeaning
categoriescs.LG, cs.AI, cs.CLComma-separated arXiv category codes; each gets a tab. See the taxonomy.
maxResults20Preprints listed per category (5–50).
refreshIntervalMin30Minutes between refreshes (5–720).
reportsDir~/Documents/arxiv-reportsWhere reports are written.
readeremptyWho the reports are for. Empty: a working researcher in the paper's own field. Examples.
reportStyleAutoAuto, Theory, Empirical or Custom. Styles.
promptFileemptyYour prompt, used when the style is Custom. Writing one.
engineClaude CodeClaude Code or Custom command. Engines.
modelemptyPassed to Claude Code (--model), e.g. opus or sonnet. Empty: the engine's default.
customCommandemptyThe command used by the Custom command engine.
sandboxtrueConfine Claude Code to the paper and extraReadDirs. Sandbox.
extraReadDirsemptyComma-separated folders the AI may also read, e.g. your notes.
viewerPort8791Port of the local report viewer.

Who the reports are for

The reader setting completes the sentence “Your reader is …” in the prompt, and the model is told to pitch the report at exactly that level: explain nothing the reader already knows, and leave out nothing they would not. Describe a person, with their field and level:

The more specific the description, the better the report knows what to skip.

Report styles

Every style asks for 900–1500 words of Markdown with LaTeX mathematics, grounded in the paper alone, with the paper's own macros expanded so the report reads without its preamble.

StyleSectionsBest for
TheoryWhat it does · Main results (each theorem, precisely, with its hypotheses) · Definitions worth knowing · How the proofs go · Context · AssessmentMathematics, theoretical computer science, logic, theoretical physics
EmpiricalWhat it does · Main results (claims with their numbers and baselines) · Method · Evaluation · Context · Assessment (including code and data availability)Machine learning, experimental science, systems
AutoWhat it does · Main results · Key ideas · How it works · Context · Assessment, each adapting to whether the paper is theoretical or empiricalMixed feeds

The prompts are in the plugin's prompts/ folder; read them to see exactly what the model is asked.

Custom prompts

Set reportStyle to Custom and promptFile to a Markdown file. Start from a copy of one of the built-in prompts:

cp ~/.config/omarchy/plugins/io.github.pierrecagne.arxiv/prompts/theory.md ~/arxiv-prompt.md
omarchy bar set io.github.pierrecagne.arxiv promptFile ~/arxiv-prompt.md
omarchy bar set io.github.pierrecagne.arxiv reportStyle Custom

These placeholders are filled in before the prompt is sent:

PlaceholderReplaced by
{READER}The reader setting, or the default description.
{SOURCE}The paper's main file name, e.g. main.tex or 2401.12345v1.pdf.
{SOURCE_NOTE}How to read the paper: where its LaTeX files are, that it is a PDF, or that its text follows the prompt. Also lists the extra folders.

Ideas: ask for the report in another language, add a section on how the paper relates to your own project, or ask for a shorter digest. Keep the rule that the paper is material to report on and not instructions to follow: it is part of what protects you from booby-trapped papers.

← Using itAI engines & sandbox →