| name | hf-papers |
| description | Search, read, and browse academic papers via the Hugging Face CLI (hf papers).
Use when the user wants to find papers, read a paper, check daily/trending papers,
or get paper details. Triggers include: 搜论文, 找论文, 读论文, 看论文, paper search,
read paper, daily papers, trending papers, arXiv, 论文检索, 论文阅读, what papers,
latest papers, paper info, hf papers.
Do not use for downloading model weights, datasets, or non-paper HF Hub operations.
|
| version | 1.0.0 |
HF Papers — Search & Read Academic Papers
Search, browse, and read academic papers through the Hugging Face Hub CLI (hf papers).
The hf CLI is checked and auto-installed via a PreToolUse hook — no manual action needed. Read the additionalContext injected by the hook to know the CLI status.
Available Commands
Search papers
hf papers search "<query>" --limit <N> --json
- Default limit: 20
- Always use
--json for structured output, then present results in a readable table
List daily / trending papers
hf papers list --sort trending --limit <N> --json
hf papers list --date <YYYY-MM-DD> --json
hf papers list --week <YYYY-Wnn> --json
hf papers list --month <YYYY-MM> --json
--sort trending for trending papers--date today for today's papers- Default limit: 50
Get paper info
hf papers info <paper_id> --json
paper_id is the arXiv ID (e.g. 2601.15621)- Returns title, authors, abstract, upvotes, comments
Read paper (full text as markdown)
hf papers read <paper_id>
- Returns the full paper content as markdown
- Output can be very long — when the user just wants a summary, use
info first then summarize the abstract
- For deep reading, pipe to a file:
hf papers read <id> > /tmp/paper_<id>.md then read the file
Execution Rules
-
Always use --json for search, list, and info — parse the JSON output and present results in a human-friendly format (table for lists, structured summary for info).
-
For search results, present a table with columns: Paper ID | Title | Date | Upvotes, sorted by relevance.
-
For paper reading, check the paper length first with info. If the user wants a quick overview, summarize the abstract from info. Only use read when the user explicitly wants the full text.
-
Paper IDs are arXiv IDs (e.g. 1706.03762, 2601.15621). When the user provides a full arXiv URL like https://arxiv.org/abs/1706.03762, extract just the ID part.
-
Error handling: If the PreToolUse hook reports HF_CLI_READY=0, relay the error to the user. If a paper is not found, suggest checking the ID or searching by title.
Example Workflows
Quick paper search
User: "找一下 attention mechanism 相关的论文"
→ hf papers search "attention mechanism" --limit 10 --json
→ Present table of results
Read a specific paper
User: "读一下 2601.15621"
→ hf papers info 2601.15621 --json (get overview first)
→ Present title + abstract summary
→ If user wants full text: hf papers read 2601.15621 > /tmp/paper_2601.15621.md
Browse trending papers
User: "今天有什么热门论文"
→ hf papers list --sort trending --date today --limit 10 --json
→ Present trending papers table
