renocrypt/latex-arxiv-skill

Overview

This skill generates IEEEtran-format LaTeX review papers for ML/AI destined for arXiv, producing a scaffolded project, verified BibTeX, and compiled outputs. It enforces a gated, issue-driven workflow that ensures citation verification, layout hygiene, and reproducible project structure. The skill is optimized for 6–10 page review articles with rigorous citation validation.

How this skill works

Start by providing a topic and optional constraints (venue, page limit, authors). The skill scaffolds an IEEEtran two-column project, runs a light literature discovery, proposes titles and an outline, and creates a plan and issues CSV that become the execution contract. Writing proceeds per-issue: research, write with citations, verify BibTeX entries via the arXiv registry, compile, and refine until QA gates pass and final artifacts (main.tex, ref.bib, main.pdf, figures, notes) are produced.

When to use it

• Preparing an ML/AI literature review for arXiv with a LaTeX+BibTeX workflow.
• Scaffolding a reproducible IEEEtran two-column review paper with verified citations.
• Validating and repairing BibTeX citations in an existing LaTeX project.
• Generating a tracked plan and per-section issues for collaborative writing.
• When you need strict layout hygiene (no Overfull \hbox) and reproducible compilation.

Best practices

• Always approve the kickoff plan before adding prose to main.tex; plan + issues are mandatory gates.
• Verify every citation via the arXiv registry or source page before adding to ref.bib; never fabricate sources.
• Split or insert new issues if scope grows; keep the issues CSV as the single source of truth.
• Compile frequently and fix Overfull \hbox warnings before marking issues DONE.
• Prefer single-column figures and escalate to double-column only when necessary for readability.

Example use cases

• Kickoff a 7-page arXiv review on foundation models: scaffold project, propose titles, and plan visuals.
• Audit and repair an existing LaTeX paper's ref.bib and produce a verification report plus corrected entries.
• Run a section-by-section writing loop: research 8–12 papers per section, add prose with citations, and mark issues DONE.
• Produce a final QA-checked PDF and LaTeX sources with 100% verified citations and 5+ visualization types.
• Generate literature notes and an arXiv metadata cache to speed future reviews on related topics.