Access and Summarize AI Research Papers

This skill looks up and reads Hugging Face paper pages (hf.co/papers) in markdown, and uses the Papers API for structured metadata such as authors, linked models, datasets, Spaces, GitHub repo, and project page. Hugging Face Paper pages are described as a platform built on top of arXiv specifically for AI and computer science research, where authors can claim their paper, link associated model checkpoints, datasets, and Spaces by referencing the paper's HF or arXiv URL in a model card, dataset card, or Space README, link a GitHub repository or project page, and link an HF organization. Papers mentioned this way are automatically indexed, though only some are also submitted to the Daily Papers feed, and daily-paper submission closes 14 days after the arXiv publication date.

The skill starts by parsing a paper ID out of whatever the user provides: a Hugging Face paper URL, its .md variant, an arXiv abs or pdf URL, a bare arXiv ID, or an ID with a version suffix like 2602.08025v1, all resolving to the same paper ID for use in the API calls below.

To fetch a paper as markdown:

curl -s "https://huggingface.co/papers/{PAPER_ID}.md"

This relies on the HTML version of the paper at arxiv.org/html; if that HTML version doesn't exist, the content falls back to the HTML of the Hugging Face paper page itself, and a 404 means the paper is not yet indexed. Markdown can also be requested from the normal page URL with an Accept: text/markdown header.

The Papers API endpoints, all under https://huggingface.co, include: GET /api/papers/{PAPER_ID} for structured metadata (authors, media, summary, AI-generated summary, project page, GitHub repo, organization, upvotes); filtering models, datasets, or Spaces by arxiv:{PAPER_ID}; POST /api/settings/papers/claim to claim authorship for a target user, given a claimAuthorId and targetUserId; GET /api/daily_papers with page, limit, date/week/month, submitter, and sort parameters; GET /api/papers for a cursor-paginated list sorted by published date; GET /api/papers/search for hybrid semantic and full-text search over title, authors, and content, with a query up to 250 characters; POST /api/papers/index to insert a paper from arXiv by ID (re-indexing an already-indexed paper is restricted to its authors); and POST /api/papers/{PAPER_OBJECT_ID}/links to update the project page, GitHub repository, or submitting organization, restricted to the paper's author, the Daily Papers submitter, or a papers admin.

Error handling notes a 404 on the page, .md, or /api/papers/{PAPER_ID} endpoint as not-yet-indexed, and recommends verifying the extracted arXiv ID including any version suffix if a paper isn't found. If the Hugging Face paper page lacks enough detail, the fallback is the regular paper page and then the original arXiv abs or pdf URL.

Use it when a user shares a Hugging Face paper page URL, its markdown variant, an arXiv abs or pdf URL, or a bare arXiv ID, or asks to summarize, explain, or analyze an AI research paper.

Its own limitations say to use it only when the task clearly matches this upstream product's scope, and to verify commands, API behavior, pricing, quotas, credentials, and deployment effects against current official documentation before making changes.

Inputs are a Hugging Face paper URL, arXiv URL, or arXiv ID, and for write operations an HF_TOKEN bearer token. Outputs are the paper's markdown content, structured JSON metadata, lists of linked models/datasets/Spaces, Daily Papers feed results, search results, or a confirmation of a claim, index, or link update.

It connects to the Hugging Face Hub API and, indirectly, to arXiv for the HTML source of paper content. No authentication is required for public read endpoints; claim authorship, index paper, and update paper links all require an Authorization: Bearer $HF_TOKEN header. The .md endpoint is preferred for reliable machine-readable output, and /api/papers/{PAPER_ID} is preferred when structured JSON fields are needed instead of page markdown.

It is for researchers, developers, and agents that need to read, search, or manage metadata for AI research papers indexed on Hugging Face, including authors managing their own paper's links and claims.