Academic Research Skills: a 39-agent research team inside Claude Code

The problem

Finding solid references and formatting citations eats hours, and most researchers do it alone. The failure mode is usually not the writing—it's letting an unverified reference or an ungrounded claim slip into a draft that took days to produce.

Academic Research Skills is an open-source suite of Claude Code skills (41K+ stars, CC-BY-NC 4.0) that wires a collaborative team of specialized agents around that workflow, from a first research question to a compiled PDF.

What the suite is

It ships as four skills that work independently or as one pipeline:

• deep-research (14 agents) — question formulation, literature search, source verification, meta-analysis, and synthesis.
• academic-paper (12 agents) — structure, argument building, drafting, citation compliance, and formatting across five citation styles (APA, IEEE, Chicago, MLA, Vancouver).
• academic-paper-reviewer (7 agents) — a multi-perspective peer-review panel.
• academic-pipeline (5 agents) — the orchestrator connecting the other three.

In total that's a claimed ~39-agent ensemble across 27 modes, routed by how you ask.

The routing model

You don't pick a mode explicitly—you describe the task and Claude Code routes it:

You say	Claude runs
"I have a vague idea, help me frame it"	deep-research socratic (guided dialogue)
"Give me a literature summary on X"	deep-research quick
"Help me write a paper about Y"	academic-paper full
"Review this paper"	academic-paper-reviewer full
"Produce a complete research paper on Z"	academic-pipeline (full 10-stage run)

The socratic modes are the interesting part: instead of handing back an answer, the mentor agent asks 5–15 rounds of questions to force you to sharpen the research question yourself.

The full pipeline

academic-pipeline chains the other skills into a 10-stage run:

1. Research — question refinement (FINER-style), literature search, source verification.
2. Write — structure architect builds an outline with word allocation; argument builder maps thesis and evidence chains.
3. Integrity check (Stage 2.5) — every reference, data point, and claim is audited before review. This is the guard that catches hallucinated citations.
4. Peer review — editor-in-chief plus domain reviewers and a devil's advocate score the paper on a 0–100 rubric.
5. Coaching — a Socratic pass explains why points were raised, so you learn while revising.
6. Revise — each review comment is tracked to a resolution status.
7. Re-review (Stage 3′ with integrity gate 4.5) — panel checks the revised paper.
8. Re-revise — address any remaining points (capped loops).
9. Final integrity check — rerun the citation/claim audit on the final version.
10. Finalize — format conversion and output in Markdown, DOCX, and LaTeX → PDF (Tectonic).

The pipeline is built around mandatory integrity gates between research, review, and finalization, so a claim can't reach the finished PDF without being checked.

Installation

The repo supports several install paths (see SETUP.md). The quickest project-local setup:

git clone https://github.com/Imbad0202/academic-research-skills.git ~/academic-research-skills
cd /path/to/your/project
mkdir -p .claude/skills
ln -s ~/academic-research-skills/deep-research .claude/skills/deep-research
ln -s ~/academic-research-skills/academic-paper .claude/skills/academic-paper
ln -s ~/academic-research-skills/academic-paper-reviewer .claude/skills/academic-paper-reviewer
ln -s ~/academic-research-skills/academic-pipeline .claude/skills/academic-pipeline

Then run claude in that directory. You can also install globally under ~/.claude/skills/ or via the Claude Code plugin marketplace.

Practical notes

• Start with one skill. A full pipeline run is expensive—the README estimates roughly $4–6 in API cost and 2–4 hours of collaborative work. deep-research alone or the reviewer alone is a low-cost way to test before committing token budget.
• The license is non-commercial. It's CC-BY-NC 4.0, not MIT—fine for personal and academic use, but check before using it to drive revenue-generating work.
• Multi-agent stages want subagents enabled. The repo recommends agent teams for the parallel research/writing/review stages.
• Claude Code skills are loaded from SKILL.md. Each of the four directories must sit at .claude/skills/<skill-name>/SKILL.md for discovery to work.

Takeaway

The highest-value idea here isn't the number of agents—it's the mandatory integrity gates. Running a citation and claim audit before the paper goes to review (and again before finalization) is exactly the step most writers skip. Even if you never run the full 10-stage pipeline, borrowing that single habit—verify every reference before you let a draft be "finished"— improves any research workflow.

---

Source: the academic-research-skills repository and its README. Star count and pipeline details verified from the repository on 2026-08-07.