Stop fake citations: Academic Research Skills for Claude Code

Researchers found 146,932 fake citations in papers from 2025 alone. Academic Research Skills is a free (non-commercial) Claude Code plugin that checks every reference in your paper exists, can check that each source says what you claim, and runs a mock peer review with a devil's advocate. It still misses some, so read your key sources yourself. This is the full version of the CITE guide from my Reel. Facts checked on 20 Oct 2026 against the project README and the arXiv paper linked below.

01What it is

Academic Research Skills (ARS) is an open project by an independent developer (GitHub user Imbad0202). It is a set of Claude Code skills for academic work: research, writing, review, revision and final formatting. Three parts matter most for citations:

Source check. Deep Research verifies references through the Semantic Scholar API, and a deterministic citation gate checks that each reference resolves to a real work.
Claim audit (opt-in). Turn it on with ARS_CLAIM_AUDIT=1. It fetches the cited source and judges whether it actually supports the sentence that cites it. Problems such as claim-not-supported or fabricated-reference block the output.
Reviewer panel. The Academic Paper Reviewer runs a Journal-Fit Reviewer, three more reviewers and a Devil's Advocate against your draft.

The pipeline stops for your confirmation at every stage. You stay the author.

02Why it matters

Zhao et al. (arXiv 2605.07723, May 2026) audited 111 million references across 2.5 million papers on arXiv, bioRxiv, SSRN and PubMed Central. Their conservative estimate is 146,932 hallucinated (non-existent) citations in 2025 alone. They also found that preprint moderation and journal publishing catch only a fraction of them. A reference that looks right is not proof that it exists.

03Setup, step by step

Install the latest Claude Code (plugins need a recent version).
Make sure an Anthropic API key is available: export ANTHROPIC_API_KEY, or set it the first time you run claude.
In Claude Code, add the marketplace and install the plugin (commands below).
Check that it works: run /ars-plan and describe the paper you are working on. It starts a guided conversation to map out your structure. For a quick single test, try /ars-lit-review "your topic".
To check that sources really support your claims, set the environment variable ARS_CLAIM_AUDIT=1 before you start Claude Code.
Optional: Pandoc for DOCX output, and tectonic for APA 7.0 PDF. Markdown output works without either. Other install methods (project skills, global skills, a claude.ai Project, Claude Science import) are in docs/SETUP.md.

Install in Claude Code

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

04Good to know

The claim check is off by default. Without ARS_CLAIM_AUDIT=1 you get existence checks, not "does this source say that" checks.
Non-commercial licence. ARS is CC BY-NC 4.0. You can use and adapt it for free for non-commercial work. Commercial use needs the author's permission.
Data leaves your machine. Bibliographic lookups go to outside services. docs/DATA_FLOWS.md lists what is sent and how to turn each path off.
It is not a humanizer. It will not hide that you used AI. Follow your school's or journal's AI rules.