CLAIM Checklist Skill for Language Models
The CLAIM checklist skill is an Agent Skill that guides a large language model (LLM), such as Claude, through the 44 items of the CLAIM 2024 checklist for a medical imaging AI manuscript. It works in two modes:
- Author mode — completes the checklist for your own manuscript, with page/line locations for each Yes, a reason for each No or NA, and suggested text for fixable gaps.
- Assessor mode — rates someone else’s manuscript (peer review, meta-research), with a quote or section pointer as evidence for each answer.
The output is a checklist table, Yes/No/NA totals, and a list of borderline items for human review. The skill adds item-specific answer rules for the items that LLMs most often misjudge and refers the model to the item elaborations on this site.
Download skill (.zip) View on GitHub
Installation
Claude apps (claude.ai, desktop)
Download the skill archive above and upload it in the Skills section of the Claude settings.
Claude Code
Copy the skill into your personal skills folder (or into .claude/skills/ of a project):
mkdir -p ~/.claude/skills/claim-checklist
curl -fsSL https://raw.githubusercontent.com/RSNA/claim/main/skills/claim-checklist/SKILL.md -o ~/.claude/skills/claim-checklist/SKILL.md
Other agents and language models
Tools that support the Agent Skills format load the unzipped claim-checklist folder from their skills directory. For any other LLM, paste the contents of SKILL.md (without the front matter between the --- lines) at the start of the conversation.
Usage
Provide the complete manuscript, including tables, figure legends and supplements, and ask, for example:
- “Fill out the CLAIM checklist for my manuscript.”
- “Assess this paper’s adherence to CLAIM 2024.”
- “Which CLAIM items does this draft still miss?”
The skill is licensed under MIT. The CLAIM item text it contains is © RSNA and reproduced by permission; it is not covered by the MIT grant.