Searching the Literature

You can search the literature in two ways. The /scholar command runs a guided research session from topic to written summary. You can also ask Claude in plain words, for example “find papers on psychological safety in hybrid teams”, and the researching-topics skill takes over.

Start a Research Session

1. Type /scholar followed by your topic, or type /scholar alone and Claude asks what you want to research.
2. Answer the three scope questions Claude asks about depth, databases and dates.
3. Review the table of results Claude presents, and say which papers you want to explore or keep.
4. Read the summary Claude writes, then say whether to import any of the papers to your library.

The three scope questions set the shape of the search.

Question	Choices
Depth	A quick scan of the top 5 results, or a comprehensive search of 15 or more
Databases	All ten, or the ones you name
Date range	Any date, the last few years, or newest first since a date you give

What Happens During a Session

The researching-topics skill guides Claude through a fixed sequence of steps.

1. Claude writes two or three versions of your query, using synonyms and related terms to cover different angles.
2. It runs scholar_search with each version and checks your library with library_search for papers you already hold.
3. It removes duplicates across the result sets and ranks what is left by citation count, recency and journal quality.
4. It shows the top 10 to 15 results in a table and asks which papers you want to explore further.
5. For the papers you choose, it fetches full details, full text where available and related papers through their citations.
6. It offers to import the chosen papers to your library so they are there for later sessions.
7. It writes a summary organised by theme, with APA7 citations and a confidence tag on each theme.

Claude saves the summary at the location you give. Without one, the skill names the folder ~/Documents/Agent Outputs/ followed by a project identifier.

Confidence Tags

Each theme in a summary carries one of three tags, which describe the strength of the evidence behind it.

Tag	Meaning
HIGH	Several replicated studies, meta-analyses or systematic reviews with consistent findings
MEDIUM	One well-designed study, limited replication, or findings from a narrow population
LOW	Untested theory, anecdotal evidence, unresolved conflicting findings, pilot studies or small samples

The skill never gives HIGH to a single study and caps unreplicated preprints at MEDIUM. When the evidence conflicts but both sides are well supported, it notes the disagreement and uses MEDIUM.

The Search Wizard

The search wizard, the scholar_wizard tool, reads your query before a search. Its tool description asks Claude to call it before searching, but /scholar and the researching-topics skill do not call it as a fixed step, so ask for the wizard by name when you want it. It detects the discipline from keywords, suggests databases for that discipline and gives Claude a short list of questions to ask you.

• Discipline: asked when no keyword in your query points to a field, or only one keyword does.
• Access: open access only, or gated papers included as secondary results. The wizard’s suggested answer is open access only.
• Date range: the last 5, 10 or 20 years, or any date. The suggested answer is the last 10 years.
• Preprints: asked for neuroscience, biomedical, computer science, mathematics and physics, economics and multidisciplinary queries.
• Authors: an optional free-text question about authors to prioritise.

The Data Sources page lists which databases the wizard suggests for each discipline.

Search Options

Claude chooses these options for you during a session, and you can ask for any of them by name or in plain words.

Option	What It Does
query	The words to search for
sources	The databases to search. All ten by default, with CORE skipped when you have no key
max_results	The most results to take from each database. The default is 10, and the total can be higher before duplicates are merged
open_access_only	Keeps only open access papers when set to true. By default gated papers stay in, listed after open access ones
year_from, year_to	The earliest and latest publication year
published_after, published_before	An exact date window, as YYYY-MM-DD, YYYY-MM or YYYY. Finer than the years, and preferred over them
sort_by	The order of results, from the four modes below
venues	Keeps only papers whose journal or publisher name contains one of the names you give, ignoring capitals, for example Journal of Applied Psychology or NBER
exclude_dois	Drops papers with the DOIs you list, for example papers already in your library
citation_style	Returns one citation style for each paper. By default all six styles come back

The four sort modes change which papers you see first.

Mode	Order
open_access	The default. Open access papers first, then by citation count
citations	Most cited first
date	Newest first, which suits a search for recent papers
relevance	Each database’s own ranking, taking turns between databases

Note

The venues filter works on the results after they arrive, so it can only keep papers the databases already returned. Raise max_results when you filter by venue, and put the journal or publisher name in the query as well.

Reading the Results

Each paper in the results carries its title, authors, abstract, year, journal, DOI and a link. It also shows which database returned it, its citation count where the database gives one, and its publication date where known. Three fields describe access.

Field	Meaning
openAccess	Whether the paper is openly available
openAccessUrl	A direct link to the open access version, where there is one
fullTextAvailable	Whether PRIMA Scholar can fetch the content or a download link

The response as a whole reports the number of open access and gated papers, the databases that answered and the sort order used. It also lists any filters in force. Two more parts appear only when needed, errors for a database that failed and missingApiKeys for a database skipped for want of a key.

Finding What Is New Since a Date

To see what has appeared since a given date, ask for papers published after that date in newest-first order. Claude then sets published_after to your date and sort_by to date. New papers have few citations, so the default order would push them down the list.

With a date sort, PRIMA Scholar caps the window at today unless you give your own end date. This keeps out records that some publishers list with placeholder dates in the future. To skip papers you already hold, ask Claude to pass their DOIs as exclude_dois.

For a scan you repeat on a schedule, use the Literature Watch instead.

Following Citations

The scholar_citations tool follows a paper’s links in the Semantic Scholar citation graph. Ask for the papers that cite a known paper to find newer work built on it. Ask for the papers it cites to find the earlier work it rests on.

You can name the paper by Semantic Scholar ID, or by a DOI, arXiv ID or PubMed ID with its prefix. Claude returns up to 10 papers unless you ask for more.

Looking Up One Paper

The scholar_get_paper tool fetches full details for one paper. It accepts a DOI, a PubMed ID such as PMID:38123456, an arXiv ID such as ARXIV:2301.12345 and an OpenAlex ID such as W1234567890. It also takes ERIC numbers, PMC IDs and Semantic Scholar paper IDs. A DOI lookup tries Semantic Scholar first and falls back to CrossRef, while CORE IDs and DBLP keys do not work, so use the DOI for those papers.

Getting the Full Text

The scholar_full_text tool tries four routes in order and stops at the first that works.

1. CORE: with a CORE key, it looks the ID up as a CORE work and returns the text itself when CORE holds it.
2. Europe PMC: for a PMC ID, PubMed ID or DOI, it returns the Europe PMC article page link when Europe PMC lists the paper as open access and holds it under a PMC ID.
3. arXiv: for an arXiv ID, it returns the arXiv PDF link, built from the ID without checking that the paper exists.
4. bioRxiv/medRxiv: for a DOI that starts with 10.1101/, it returns the https://doi.org/ link.

CORE is the only route that returns the text itself. When no route applies, the tool says the paper may be behind a paywall and suggests checking institutional access or the publisher’s site.

Writing Better Queries

The researching-topics skill follows a few habits that you can use in your own requests.

• Start broad, then narrow: name the two or three core ideas in your question and list synonyms for each.
• Use precise phrases: name the established term for the idea, such as self-determination theory.
• Combine ideas: “attachment theory workplace” narrows better than either idea alone.
• Add method words: “systematic review” or “meta-analysis” pulls in stronger evidence when you want it.
• Widen when results are thin: remove date limits, try broader terms or search a different database.