Academic Research Skills

Why use it

Key features

5 stage skills with handoffs (research → write → review → revise → finalize)
Citation-aware (BibTeX / arxiv IDs)
Outline-first drafting — no AI-blob prose
Self-review prompts the model to challenge itself
LaTeX + Markdown output
MIT, no telemetry

Live Demo

What it looks like in practice

ready

Install

Author repository · Claude Skills documentation

In Claude Code, install the plugin using the author marketplace commands:

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

Run /ars-plan in Claude Code and describe your paper topic. Follow the author setup guide for prerequisites and other installation channels.

Use Cases

Real-world ways to use Academic Research Skills

Build a literature review from a blank doc to first draft

👤 Grad students starting a chapter ⏱ ~180 min intermediate

When to use: You have a topic and the dread of starting.

Flow

Research

Use the research skill. Topic: 'multi-agent LLM systems for code'. Find 25 recent papers, categorize by approach, save bibliography.✓ Copied

→ Categorized list of 25 papers, .bib file written
Outline

Now use the write skill. Build an outline that organizes those papers into a coherent narrative.✓ Copied

→ Outline with section heads + which papers go where
Draft

Draft section 2 using only the papers in your outline. Cite inline with \cite{}.✓ Copied

→ Section drafted; citations resolve

Outcome: First draft of the lit review with proper citations, ready for revision.

Pitfalls

Research skill returns papers Claude hasn't actually read — Pair with arxiv-mcp-server so abstracts and full PDFs are pulled, not summarized from titles

Combine with: arxiv-mcp-server

Run the review skill on your own draft before showing your advisor

👤 Anyone wanting a tougher first read ⏱ ~60 min intermediate

When to use: You're done; you suspect rough patches you stopped seeing.

Flow

Adversarial review

Use the review skill. Read /paper/main.tex as a skeptical reviewer. Identify weakest 3 claims, missing citations, structural issues.✓ Copied

→ Per-claim critique
Revise

Use the revise skill to address the top 3 review items. Don't change anything else.✓ Copied

→ Targeted edits

Outcome: A draft that survives advisor review.

Pitfalls

Reviewer is too gentle — Pass --style harsh so the prompt encodes a tougher persona

Combine with: paperdebugger-mcp

Finalize for a specific journal/conference template

👤 Authors at the submission stage ⏱ ~30 min intermediate

When to use: You have a draft; the venue uses a specific LaTeX class.

Flow

Apply template

Use the finalize skill. Wrap /paper/main.tex into the IEEEtran template. Map abstract, sections, figures.✓ Copied

→ Templated source compiles
Compliance check

Verify formatting against the venue's policy (page limit, font sizes, margins).✓ Copied

→ Compliance report; flags if any

Outcome: Submittable document.

Pitfalls

Some venues require their own bibliography style — Skill checks .bst files and tells you if missing

Combinations

Pair with other MCPs for X10 leverage

academic-research-skills-imbad + arxiv-mcp-server

Pull real abstracts and PDFs to ground research

When researching, fetch full PDFs for top 10 papers via arxiv MCP.✓ Copied

academic-research-skills-imbad + paperdebugger-mcp

Use PaperDebugger's multi-agent reviewer for a tougher review

After self-review, run PaperDebugger reviewer for a second opinion.✓ Copied

Tools

What this MCP exposes

Tool	Inputs	When to call
research	topic, depth?, max_papers?	Stage 1
write	outline, papers, target_length	Stage 2 — drafting
review	draft_path, style?	Stage 3
revise	draft, review_items[]	Stage 4
finalize	draft_path, template	Stage 5

Cost & Limits

What this costs to run

API quota: Bound by LLM
Tokens per call: Lit review can be 50k+ tokens; full paper 100k+
Monetary: Free; LLM cost is yours
Tip: Stage by stage — don't try to do all five in one call

Security

Permissions, secrets, blast radius

Minimum scopes: filesystem-read filesystem-write

Credential storage: None

Data egress: LLM provider; arxiv if combined

AI-assisted writing rules vary by institution — check yours

Troubleshooting

Common errors and fixes

Citations are fabricated

Always pair with arxiv MCP or another real-source fetcher; never let research run on title-only

Verify: Check each citation on Semantic Scholar

Drafts feel formulaic

Pass author voice samples via --voice voice.md

Finalize template fails to compile

Confirm \usepackage and template files are in the project root

Verify: `pdflatex` or `latexmk` runs cleanly

Alternatives