Academic Research Skills for Claude Code

# Academic Research Skills for Claude Code

Claude Code skills that guide academic research from planning to publication, with your confirmation at every stage, integrity checks and an optional citation audit.

Made by imbad0202 on GitHub. These notes are Mindket's, written from its README; the project's own files and license are in its repository.

## About

What it is: Academic Research Skills is a set of Claude Code skills that covers academic work from first plan to publication. Claude drafts, finds references, formats citations, checks data and tests logical consistency, while you make the decisions. The pipeline pauses for your confirmation at every stage, and integrity checks and an optional citation audit guard against invented results and sources.

Who it suits: It suits researchers and students who write papers, review them or run literature reviews, and who want AI help without handing over authorship. You stay the author and answer for every claim you submit.

```table
Industry | Who uses it | How they use it | Example
Academia | Graduate students and researchers | Shape a paper's structure through guided questions before drafting | Run /ars-plan on a paper idea
Academic writing | Authors drafting papers | Match their own voice and catch prose that reads as machine-written | Use Style Calibration on past work
Peer review | Reviewers | Use a reviewer that can measure its own error rates against a gold set you supply | Calibrate the reviewer on your labeled examples
Research integrity | Anyone submitting under their own name | Check that each cited source supports the claim it is attached to | Turn on the optional claim audit
```

If you want AI use hidden, or a paper written and submitted with no human in the loop, skip it. It is built for the opposite.

Where to use it:

- Claude Code in the terminal, in VS Code and in JetBrains, installed as a plugin.
- The Claude web app and Claude Desktop on paid plans, by adding the same repository as a plugin marketplace.
- Hooks and agents that ship with a plugin stay idle in ordinary chats there.

```table
Use case | Who | What they get back
Paper planning | Authors starting a paper | A paper structure worked out through back-and-forth questions
Drafting help | Authors | Draft text, hunted-down references, formatted citations and checked data
Integrity gates | Anyone using AI drafts | A blocking checklist of seven AI research failure modes at two points in the pipeline
Claim audit | Authors and reviewers | A check that each cited source supports its claim, with unsupported ones refused at output
```

How to set it up:

- Claude Code in the terminal, VS Code or JetBrains. The README gives v3.7.0 or later for this route.
- For the web route, a paid Claude plan on the web app or Claude Desktop.

1. In Claude Code, add the marketplace with the first command below.
2. Install the plugin with the second command.
3. Run /ars-plan to start working through your paper.
4. On the web app or Claude Desktop, open Customize, then Plugins, then Add, then Add marketplace, then Add from a repository. Enter the same repository name and install academic-research-skills.

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

The README also has a Quick install section with prerequisites and a traditional symlink route.

How to use it: Start with /ars-plan, which asks you questions one after another to shape your paper's structure. From there the pipeline moves stage by stage and stops for your confirmation each time. To turn on the optional claim audit that arrived in v3.8, use this setting:

```text
ARS_CLAIM_AUDIT=1
```

```table
Tool | What it does
/ars-plan | Walks you through your paper's structure with guided questions
Style Calibration | Learns your voice from your past work
Writing Quality Check | Flags patterns that make prose read as machine-written
Integrity gates | Run a seven-mode blocking checklist at Stage 2.5 and Stage 4.5
ARS_CLAIM_AUDIT=1 | Fetches each cited source and judges whether it supports the claim
```

Example: You could run /ars-plan on a half-formed paper idea. It asks about your question and approach in turn and helps settle the structure. The pipeline then waits for your confirmation before it moves to the next stage.

For newcomers and experts: Think of it as a set of helpers for the long road from idea to submission, with you as the author. A good first try is /ars-plan, which only asks questions and needs no setup beyond the install.

Under the hood, the project rests on the idea that a researcher working with AI avoids autonomous-AI failure modes better than a human or an AI alone. Since v3.7.3 each citation carries a three-layer locator anchor. In v3.8 the opt-in audit fetches the source for each anchor and judges the claim, and five new high-warning classes make the formatter refuse output. The trade-off is that the shipped calibration test uses a stub judge returning the gold labels, so no live-judge result exists yet.

```flow
Each citation carries a locator anchor
The audit fetches the source for each anchor
A judge decides if the claim is supported
Unsupported claims raise high-warning flags
The formatter refuses output that trips them
```

Good to know:

- The README badge gives the license as CC BY-NC 4.0, which is non-commercial.
- In full mode the AI can draft a whole paper, but the pipeline still stops for your confirmation.
- The claim audit is opt-in and fetches cited sources, so it needs access to them.
- The calibration runner brings its own made-up reference set of 25 tuples, and passing means a false-negative rate under 0.15 and a false-positive rate under 0.10.
- A live-judge calibration result is not recorded yet, and testing ARS across a whole corpus is future work.
- The README cites an audit that estimated 146,932 hallucinated citations for 2025 alone, which is the problem its citation checks address.
- The latest version in the README is v3.24.2, and the README is also offered in several other languages.

## On GitHub, 15 November 2026

- Stars: 56,904
- Forks: 4,339
- Watching: 146
- Open issues and pull requests: 45
- Language: Python
- License: see the repository
- Topics: academic-pipeline, academic-writing, ai-research, claude, claude-code, literature-review, peer-review, prompt-engineering

## Links

- Code: https://github.com/imbad0202/academic-research-skills
- Website: https://buymeacoffee.com/crucify020v
- Releases: https://github.com/imbad0202/academic-research-skills/releases
- Issues: https://github.com/imbad0202/academic-research-skills/issues
- Star history: https://www.star-history.com/#imbad0202/academic-research-skills&Date

## Get the code

```
git clone https://github.com/imbad0202/academic-research-skills.git
```

Then follow the setup in its README. Nothing in this kit runs by itself.

README.md  7 KB