Advanced: One Claude Skill Runs a Full Research Workflow From Literature Search to Peer Review

What You Get

An open-source academic research plugin, Academic Research Skills (hereafter ARS), installed into Claude Code. It turns the most easily missed and most mechanical parts of writing a paper into checklists you can run repeatedly: literature review and research gaps, paper outline, citation checking, simulated peer review, and AI use disclosure.

Get the division of labor straight first: ARS acts as inspector and sparring partner, and you write the body text yourself. It can indeed generate a whole paper with one click, but that is exactly the part you should not use, for the reasons at the end.

First, Understand What It Is and Isn't

It is a set of Skills plus commands: after installation, you call them in Claude Code with commands starting with /ars-, and each step stops to wait for your confirmation.
It is not an "AI rate reduction" tool: the project description is very direct. It does not help you hide the use of AI; it only helps you make your paper more solid. Authorship and responsibility for every sentence remain with you.
The license is CC BY-NC 4.0: personal study and research are fine, but it may not be used for commercial purposes.
It incurs API costs: according to the project's own estimate, running a full 15,000-word paper costs roughly 3 to 7 US dollars (excluding cache discounts). Running only some of the steps is much cheaper.

Step 1: Install in Claude Code

First confirm that Claude Code is the latest version and works properly, then enter the following in order:

/plugin marketplace add Imbad0202/academic-research-skills
/plugin install academic-research-skills

If you want to export to Word or PDF, you also need to install Pandoc (and tectonic for PDF typesetting); if you only need Markdown, you don't. People using Codex can use a separate Codex version provided by the project.

Done when: entering /ars-plan shows it starting to ask you questions.

Step 2: Define the Research Question Before Anything Else

Put your research materials (experimental results, notes, papers you have read) into the project folder, then enter /ars-plan. It will work through a question-and-answer exchange to help you clarify the research question, chapter structure, and argument route.

Do not open with "help me write a paper", because you will most likely get a draft that is correct in every detail but says nothing. The recommended order and what to say at each step:

New accounts get 10 free credits, plus 1 more for each daily check-in. Spend 1 credit to unlock and it stays open for you.

Done when: you can state your research question in one sentence, and which research gap it fills.

Step 3: Literature Review and Research Gaps

Use /ars-lit-review "your topic" to generate an annotated reading list. It will try to verify that the works exist, but you must personally open and read each one before it can go into your paper.

Done when: for every key work on the list, you can state its main conclusion.

Step 4: Build the Outline, Then Write the Body Yourself

/ars-outline provides a detailed outline and an "evidence map" (which materials support each section). After confirming the outline, you write the body text. You may ask Claude to help organize your thinking, find counterexamples, or explain concepts you don't understand, but every paragraph you submit should come from you.

Done when: every section of the outline has corresponding evidence, and the body text is written by you.

Step 5: Check Citations

After writing, run /ars-citation-check. Fabricated citations by AI are the most common and most easily caught problem in papers: the authors, journal, and year all look real, but the article does not exist.

Do not treat the check results as a guarantee. In one complete public run of the project, after three rounds of completeness checks, a separate full review still found problems with 21 of 68 references. The tool can help you narrow the scope, but checking each one individually is still your job.

Done when: you have personally confirmed that every citation exists and actually supports the sentence you wrote.

Step 6: Simulated Review, Then Revise

/ars-reviewer assembles a simulated review panel (a journal-matched referee, several field reviewers, and a dedicated devil's advocate role) to raise issues from multiple angles. After receiving the comments, you can use /ars-revision-coach to organize a revision roadmap.

Simulated review is a sparring exercise to help you spot problems early. It cannot replace genuine peer review.

Done when: you have revised according to every review comment, or written down a reason for not revising.

Step 7: Write the AI Use Disclosure

Many journals and schools require you to state which AI tools were used in writing the paper and what they were used for. /ars-disclosure will help you prepare this according to the requirements of your submission target. State it honestly and do not leave anything out.

Done when: the statement clearly names the tools used and their specific purposes, and complies with the school's or journal's policy.

Notes

/ars-full can generate a whole paper in one go; do not use it to produce the body text you will submit: the time saved is far outweighed by the cost of academic misconduct.
Tools make mistakes: citation checks and review comments can miss problems or make errors of judgment, so verify important conclusions yourself.
Read your school's and journal's AI policies first: what AI may be used for and whether disclosure is required vary by institution.
Keep your writing process: keep drafts, notes, and revision records; if questioned, this is the best proof.
Mind the license: CC BY-NC 4.0 does not permit commercial use, so do not package it as a paid service.

Academic Integrity: This Line Must Not Be Crossed

Text generated by large models such as Claude carries statistical features that can be detected (commonly called algorithmic watermarks or algorithmic fingerprints), and major vendors are also advancing text watermarking technology. Submitting AI-generated body text directly, even after changing a few sentences, is very likely to be flagged by school detection systems, with consequences that may include a zero grade, a required retake, or even a permanent academic misconduct record. For real examples and a complete "allowed / not allowed" list, see the experience Never Let Claude Write Your Thesis Directly.