screen — Claude Code prompt: paper screening and setup
This prompt for Claude Code screens papers using their titles and abstracts. It ranks candidates by how closely they relate to a research idea.
no license ★ 159 updated 2026-11-02
.claude/commands/screen.mdWhat screen does
The prompt decides whether a paper merits a full read, using only its title and abstract. It assigns each paper a high, medium, or low relevance level and gives a one-sentence reason in the requested language.
A high rating covers work on a closely related problem or method, as well as a baseline that calls for comparison. Medium marks an adjacent direction that may contribute methods, data, or counterexamples; low means the wording is related but little else is. When uncertain, the prompt favors a higher rating, since overlooking an overlapping paper is considered worse than reading one extra paper.
Provide the research idea, a restatement, and candidate papers with IDs, years, venues, titles, and abstracts. The result must cover every paper and preserve its ID exactly. Abstracts may overstate findings or omit details, so the screening does not determine whether someone has already proposed the same idea.
What it helps with
- screen papers for a full read using titles and abstracts
- sort candidate papers into three relevance levels
- surface adjacent work that may contribute methods or data
- preserve each paper ID in the results
Example prompts for the agent
- Screen these papers against my research idea using their titles and abstracts.
- Rate each paper's relevance and keep its ID unchanged.
- Return the paper screening results in the required JSON format.
How to install screen
Manually: put the files in place
.claude/commands/screen.mdHow to use
Paste the text at the start of a chat or into project instructions
Questions about screen
How do I install screen in Claude Code?
Save the prompt file at .claude/commands/screen.md.
What does screen do?
It rates papers by their relevance to a research idea using their titles and abstracts, then gives a brief reason for each rating.
What information does screen need?
Provide the research idea, a restatement, and each candidate paper's ID, year, venue, title, and abstract.