Guide · Agent skill
How to use the Literature Review Assistant agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The literature-review skill is a set of instructions, reference files and helper scripts that teaches an AI agent to run a systematic, scoping or narrative literature review. It is aimed at graduate students writing a thesis literature review, researchers preparing a systematic or scoping review, and writers who need checked citations and a formatted Markdown or PDF document. After this guide you will know what the skill does, how to install it, how to ask for it in plain language, and what it needs to run.
Literature Review AssistantK-Dense-AI/scientific-agent-skills · MIT Download the skill Quick start
- 1Download the ZIP. Get the literature-review ZIP from the skill's page in the directory.
- 2Place the skill folder. Unzip into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. Alternatively, upload the ZIP under Settings → Capabilities → Skills in the Claude apps.
- 3Install the prerequisites. Make sure Python 3.10+ with requests is available. Add Pandoc and XeLaTeX only if you need PDF output.
- 4Describe your review. State the topic, review type, databases and output format in a plain request.
- 5Check the search log. Ask the agent to record search strings, dates and result counts so the search can be reproduced.
- 6Verify and review the output. Have the agent run citation verification, then read the references and claims against the actual sources yourself.
What is an agent skill, and what does this one make the agent do?
An agent skill is a folder containing a SKILL.md file with instructions, plus optional reference files and scripts, that an AI agent such as Claude can load when a task matches the skill's description. This skill covers the literature-review task.
The skill makes the agent follow a seven-phase process: planning the question and inclusion criteria, recording search strings, screening records, extracting data, synthesizing findings by theme, verifying citations, and generating a formatted document. It also provides PRISMA-style counting guidance and a review template.
- Planning and scoping: question, inclusion and exclusion criteria, scope
- Systematic search across databases such as PubMed, arXiv, bioRxiv and Semantic Scholar, with recorded queries
- Screening in title/abstract then full-text stages, with counts kept for the PRISMA flow
- Data extraction, quality assessment, thematic synthesis
- Citation verification and document generation in Markdown and PDF
What the skill does not do
The agent still does the reading and the judgment. The bundled scripts process normalized search records, check DOI registration and render documents; they do not run a complete review automatically and do not query databases directly. The skill also does not supply a meta-analysis engine, though it supports meta-analysis planning.
Install the literature-review skill
Download the ZIP for the skill from the directory page. Then choose one of the following options.
For a personal install, unzip the folder into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
- Personal: unzip into ~/.claude/skills/
- Project only: unzip into .claude/skills/ inside your project
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Set up what the skill needs
Check the requirements before a real review. Python 3.10 or newer with the requests library is needed for DOI checks and searches, and the agent needs network access for those. PDF export needs Pandoc and XeLaTeX installed. AI-generated schematics are optional and need an OPENROUTER_API_KEY. The optional parallel-cli tool requires Parallel authentication.
- Python 3.10+ with requests
- Network access for DOI checks and optional database or web searches
- Pandoc and XeLaTeX, only if you want PDF output
- OPENROUTER_API_KEY, only for AI schematic generation
- parallel-cli with Parallel authentication, optional
Trigger the skill with a natural request
You do not need special syntax. Describe the review you want in plain language, mentioning the topic, the review type and the output you need. The skill's description covers systematic, scoping and narrative reviews, research synthesis, reproducible searches, screening, citation checking, and Markdown or PDF reviews.
Ask for the record-keeping explicitly if you need it. The skill asks the agent to record every search string and date as it goes, because a review that cannot reproduce its own search is not systematic.
- Run a systematic literature review on [topic] across PubMed and arXiv, and keep a log of every search string and date.
- Check the DOIs and citations in my draft literature review and tell me which references do not resolve.
- Draft a scoping review of [topic] with a PRISMA flow count and export it to PDF.
- Help me plan the inclusion and exclusion criteria for a review of [question] using PICO.
Generate figures and PDFs
For systematic reviews, the skill asks for the PRISMA 2020 flow template and reconciled screening counts, and recommends tables or deterministic plotting for exact numbers. Conceptual schematics are optional. The bundled generator, scripts/generate_schematic.py, needs OPENROUTER_API_KEY and creates a PNG draft; its output path must end in .png. Inspect any generated figure yourself, because automated review does not prove scientific accuracy.
- Use tables or deterministic plotting for exact numbers, effect estimates and risk-of-bias results
- Do not put fabricated study counts into an image prompt
python scripts/generate_schematic.py "Conceptual evidence map; labels from the reviewed extraction table" \
-o figures/evidence_map.png --doc-type journal --iterations 2What people use it for
- Building a reproducible search log for a thesis literature review
- Tracking PRISMA counts for records, reports and included studies in a systematic review
- Checking DOI registration and metadata in an existing bibliography
- Producing a formatted Markdown or PDF review from a structured template
Frequently asked questions
How do I install the literature-review skill?
Download the ZIP from the skill's page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. In the Claude apps you can instead upload the ZIP under Settings → Capabilities → Skills.
What do I need to run it?
Python 3.10 or newer with the requests library, and network access for DOI checks and searches. PDF export also needs Pandoc and XeLaTeX, and AI schematics need an OPENROUTER_API_KEY.
Is the literature-review skill free?
The skill is published under the MIT licence. The source does not state prices for the optional services it can use, so check those providers' terms before relying on them.
Does the skill search databases automatically?
The agent can search databases such as PubMed, arXiv, bioRxiv and Semantic Scholar, but the bundled scripts do not query databases directly. You still need to review the results and judge eligibility.
Why is my PDF export failing?
PDF rendering needs Pandoc and XeLaTeX installed. Without them the rendering step will fail, so install both or request Markdown output only.
How does this compare with doing a review by hand or with a general chatbot?
The skill adds a fixed process: recorded search strings, PRISMA count tracking, and citation checks against DOI registration. It is heavier than a casual summary, so for quick informal reading a plain chat request is usually enough.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
This skill suits graduate students who need a reproducible search log and citation checks, and researchers preparing a systematic or scoping review who must track PRISMA counts. It is overkill for quick informal summaries, and it should not be used for clinical or regulatory decisions based on its output, since automated checks do not prove scientific accuracy.
Download the skill