read_paper — Claude Code prompt: what it does, how to install
The read_paper prompt for Claude Code and ChatGPT reviews a paper in light of a research idea. It checks the abstract against the full text and rates the work by paper type.
no license ★ 159 updated 2026-11-02
.claude/commands/read_paper.mdWhat read_paper does
The prompt guides a paper review tied to a research idea. It has the model read the methods and experiments before checking the abstract, then compare each abstract claim with evidence in the paper. It separates overclaims or contradictions from limits left out of the abstract, and asks for page references such as [p.6] when citing the text.
The model identifies the paper type before applying its rating criteria. Empirical work is judged by the support for its conclusions and the quality of its experiments; surveys by coverage and synthesis, while theory is assessed through its assumptions and derivations. Author-stated limitations stay separate from observations made during review. The notes must use the required JSON format, and uncertainty should be stated rather than filled with invented details.
What it helps with
- check abstract claims against the paper's evidence
- distinguish overclaims from omitted scope limits
- rate a paper using criteria for its type
- separate author-stated limitations from reviewer observations
Example prompts for the agent
- Review this paper against my research idea: [idea and full text].
- /read_paper: assess the paper by its type and return the notes as JSON.
- Check whether the abstract's conclusions match the full text, with page references.
How to install read_paper
Manually: put the files in place
.claude/commands/read_paper.mdHow to use
Paste the text at the start of a chat or into project instructions
Questions about read_paper
How do I install read_paper in Claude Code?
Create .claude/commands/read_paper.md and put the prompt text in that file. Then invoke it with /read_paper.
What does read_paper do?
It compares an abstract with the paper's full text, records mismatches and limitations, then rates the work using criteria for its paper type. The output is JSON.
How does read_paper assess papers without experiments?
It says not to give a paper a low rating solely because it lacks experiments. The criteria depend on the paper type; theory, for instance, is judged by its assumptions, derivations and the scope of its conclusions.