A domain pipeline, not a general toolkit

Where most trending skill repos are general-purpose, this one is narrow on purpose: a suite of Claude Code skills for academic research, covering the full pipeline from research to write to review to revise to finalize. You start with something like /ars-plan, which walks through your paper’s structure via Socratic dialogue, and proceed through stages toward a finished, formatted manuscript. The specialization is the point: it encodes how an academic paper actually gets built, rather than offering generic capabilities.

The stance it leads with is equally deliberate, and it is the right one for the domain. The README is blunt: AI is your copilot, not the pilot. The tool will not write your paper. It does the grunt work, hunting references, formatting citations, verifying data, checking logical consistency, so you can focus on the question, the method, and the interpretation. That human-in-the-loop framing is not modesty, it is a design decision backed by an argument.

Why it refuses full automation

The project cites the cautionary research directly. It points to work on fully autonomous AI research systems and the failure modes they inherit: hallucinated results, shortcut reliance, methodology fabrication, and citation hallucination. Most pointedly, it cites a large-scale audit estimating on the order of 146,932 hallucinated citations across the 2025 literature, describing real citations deployed to support claims the cited references do not actually make. The whole architecture is built on the premise that a human researcher augmented by AI avoids these failure modes better than either alone.

The integrity machinery is the standout

This is where the suite earns its place over a generic writing assistant. It builds verification into the pipeline rather than trusting the model’s output:

  • Integrity gates at specific stages run a multi-mode blocking checklist for AI research failure modes, refusing to pass output that trips them.
  • Citation provenance through trust-chain frontmatter and locator anchors, so every citation carries a traceable source rather than a plausible-looking string.
  • An opt-in claim audit that fetches each cited source and judges whether the claim is actually supported, gating on classes like claim-not-supported and fabricated-reference.
  • Style Calibration that learns your voice from past work, and a Writing Quality Check that catches the patterns which read as machine-generated.

Crucially, the README distinguishes this from a humanizer: it does not help you hide that you used AI, it helps you write something defensible. That is the honest, and academically appropriate, framing.

How the later stages earn their keep

The pipeline’s back half is where the discipline shows. The review stage is not a single pass: integrity gates sit at specific checkpoints and block output that fails the failure-mode checklist, and the reviewer offers an opt-in calibration mode that measures its own false-negative and false-positive rates against a user-supplied gold set, so you can see how much to trust it rather than taking its verdict on faith. The revise stage then acts on those findings, and Style Calibration keeps the prose in your voice by learning from your past work rather than flattening everything into a generic academic register. Finalize handles formatting, with Markdown out of the box and APA-style DOCX or PDF when you add the optional tooling. The throughline is that each stage is checkable: you are not asked to trust a black box, you are handed gates and metrics you can inspect, which is the appropriate bar for work that ends up under peer review.

Install

The recommended path is the Claude Code plugin marketplace:

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

It needs a recent Claude Code and an ANTHROPIC_API_KEY. Markdown output works out of the box; for formatted DOCX or APA-style PDF, you optionally add Pandoc, and tectonic with the right fonts. Then /ars-plan starts the Socratic outline.

Where it fits, and the caveats

Reach for it if you write papers and want the mechanical, error-prone parts, reference hunting, citation formatting, consistency and citation-support checking, handled rigorously while you keep authorship. The integrity gates are genuinely differentiated; few tools treat citation fabrication as a first-class problem to verify against.

Two things to note. The license is non-standard, so review the repository terms before depending on it, and it is a solo-maintained project with donation links, young at 9 open issues as of 2026-06 but iterating quickly through versioned integrity features. Its value is real, but it is one researcher’s evolving system, not an institutional product.

It is built on the Agent Skills system documented in anthropics/skills, and for a general development methodology in the same spirit of process discipline, see superpowers. For what else is climbing, see LLM tooling, the daily digest, and the weekly report.

FAQ

Will it write my paper for me? No, by design. It handles references, formatting, verification, and consistency checks; you define the question, method, and interpretation. The README calls AI the copilot, not the pilot.

How does it handle citation hallucination? With integrity gates, trust-chain provenance, locator anchors, and an opt-in audit that fetches sources and checks whether each claim is actually supported.

Is it a humanizer that hides AI use? No. The README is explicit that it aims at quality and defensibility, not at concealing that AI was used.

How do I install it? /plugin marketplace add Imbad0202/academic-research-skills then /plugin install academic-research-skills, with an ANTHROPIC_API_KEY and optional Pandoc/tectonic for DOCX and PDF.