Arxiv Paper Writer

Use this skill whenever the user wants Claude Code to write, scaffold, compile, debug, or review an arXiv-style academic paper, especially survey papers with...

Security Scan

The skill's purpose and runtime instructions are coherent for a LaTeX paper-writing helper, but the package claims/templates referenced by its scripts and README are missing from the included files — an incoherence that should be resolved before trusting it.

Purpose & Capability

The skill's name, description, SKILL.md, and helper scripts all align with an arXiv-style LaTeX paper authoring workflow (project scaffolding, BibTeX checks, log parsing, compile/debug guidance). However, the included repository manifest and scripts (check_skill_structure.py) expect several template and asset files (templates/*.tex, templates/*.bib, templates/agent_survey_figures_tables.tex, assets/generate_paper_flowchart.png) that are referenced in README and REQUIRED_FILES but are not present in the provided file list. This is a packaging inconsistency (likely incomplete publishing) that prevents the skill from operating as described unless those files are added.

Instruction Scope

SKILL.md explicitly instructs the agent to create and edit actual .tex, .bib, and documentation files, to read LaTeX logs, and to run compile/debug loops — all appropriate for a paper-writing skill. It also asks the agent to summarize progress in Chinese and reference exact file paths. These are expected behaviors, but they imply filesystem writes and reads (project files and logs). The instructions do not ask the agent to access unrelated system credentials or remote endpoints.

Install Mechanism

There is no install spec — the skill is instruction-heavy with small helper scripts. This is low-risk from an install perspective. The included Python scripts are local and only perform static checks on files; nothing is downloaded or executed from remote URLs.

Credentials

The skill declares no required environment variables, binaries, or credentials. The referenced workflow documents suggest installing TeX Live/latexmk on the host, which is appropriate and proportional for LaTeX compilation. No unrelated secrets or external-service credentials are requested.

Persistence & Privilege

The skill is not marked always:true and does not request persistent platform privileges. It instructs file edits within the user's project directories (normal for this use case). There is no indication it would modify other skills or system-wide agent settings.

What to consider before installing

What to check before installing or running this skill: - Missing template/assets: The skill's scripts and README expect templates/ and an assets PNG that are not present in the provided file list. Ask the publisher for the missing template files (templates/arxiv_survey_main.tex, templates/full_survey_main.tex, templates/references.bib, templates/agent_survey_references.bib, templates/agent_survey_figures_tables.tex, assets/generate_paper_flowchart.png) or confirm this is an intentional minimal package. Without them, the skill cannot scaffold or reproduce the referenced workflows. - Inspect templates before use: When the missing templates are provided, review their contents for any unexpected network calls, remote-include commands, or hidden commands that could contact external servers. The current materials show no network behavior, but templates can contain arbitrary LaTeX or shell-escape usage. - Filesystem writes: The SKILL.md explicitly instructs the agent to create and edit .tex/.bib files. If you let the agent run autonomously, it will write to your working directories. Run it in a sandboxed project directory you control, not in a sensitive location. - Run helper scripts locally first: The included Python scripts (check_bibtex.py, check_latex_log.py, check_skill_structure.py) are benign static checks. Run them locally to validate your project and to see which required files are missing. Note check_skill_structure will purposely fail if templates/assets are absent. - Tooling requirements: To actually compile PDFs you will need TeX Live or MiKTeX and latexmk/bibtex as documented. Those are expected and proportional to the skill's purpose. - Autonomous invocation: The skill is user-invocable and not forced always-on. Autonomous agent invocation is allowed by default on the platform; this is normal, but if you are concerned about granting an agent write access to files, restrict execution to manual invocation or run in an isolated environment. If you want, I can list the exact missing files the packaging expects and draft a checklist of questions to ask the skill publisher to resolve the inconsistencies.

This skill guides Claude Code through an end-to-end arXiv-style paper workflow: plan the paper, initialize a LaTeX project, build a BibTeX library, write sections incrementally, create figures/tables, compile, debug, and perform a final quality review.

Core principle

Treat paper writing as an engineering loop, not a one-shot generation task. Work in small verifiable stages: plan, scaffold, cite, write, compile, inspect, repair, and review.

When starting a paper project

1. Clarify topic, paper type, target length, audience, and whether the user wants a survey, methods paper, benchmark paper, position paper, or tutorial.
2. Create or reuse a paper directory with main.tex, references.bib, figures/, sections/, and output/.
3. Use templates/arxiv_survey_main.tex for a compact starter survey, or templates/full_survey_main.tex when the user wants the richer 9-section scaffold proven in the Agent Survey experiment.
4. Read references/workflow.md for the full staged process.
5. Read only the reference files needed for the current phase.
6. When the user asks to reproduce the Paper-Write-Skill-Test workflow, use references/agent_survey_practice.md, templates/agent_survey_references.bib, and templates/agent_survey_figures_tables.tex as concrete examples.

Workflow phases

Phase 1: Plan

Create a concrete paper plan before writing prose. Include title candidates, scope, contribution claims, section outline, target references, expected figures/tables, and verification commands.

Phase 2: Scaffold

Create a compilable LaTeX skeleton first. The first milestone is a PDF that compiles even before the paper is complete.

Phase 3: Build bibliography

Construct references.bib before heavy writing. Prefer real, verifiable papers. Use stable BibTeX keys. Avoid invented citations. If uncertain about a reference, mark it for verification rather than fabricating metadata.

Phase 4: Write sections incrementally

Write one or two sections at a time. After each substantial section, compile or at least check citations and LaTeX syntax. Write Abstract last.

Phase 5: Create figures and tables

Prefer TikZ and LaTeX tables for reproducible academic artifacts. Keep figures information-dense and directly connected to claims in the text.

Phase 6: Compile and debug

Use latexmk on Linux when available. On Windows MiKTeX, use pdflatex, bibtex, pdflatex, pdflatex. Read .log files before guessing fixes.

Phase 7: Final review

Check PDF generation, citation completeness, undefined references, figure/table placement, academic tone, contribution clarity, and arXiv compatibility.

Environment guidance

• For Windows MiKTeX and Linux TeX Live basics, read references/latex_environment.md.
• For fuller Linux cloud setup and verification, read references/linux_texlive_full.md.
• For BibTeX construction, read references/bibliography.md.
• For figure and table design, read references/figures_and_tables.md and optionally reuse templates/agent_survey_figures_tables.tex.
• For final review, read references/quality_review.md.
• For reusable task prompts, read references/prompt_templates.md.

Output style

When writing project files, edit the actual .tex, .bib, and documentation files. Do not merely describe what should be written. When explaining progress to the user, summarize briefly in Chinese and reference exact file paths.