zimoliao / arxiv

arXiv Preprints

Use this skill when the task is specifically about arXiv as a source of preprints.

This is a thin source-workflow skill:

• search skill handles overall paper-search routing
• arxiv skill handles preprint discovery, browsing, fetching, and optional ingest

When to Use

Use this skill when the user wants to:

• browse recent or category-scoped preprints
• "随便看看" arXiv 上某个方向最近有什么
• search arXiv directly instead of only searching the local library
• fetch a paper from an arXiv ID, abs URL, or pdf URL
• download a preprint into the configured paper inbox (fresh default: data/spool/inbox/)
• send an arXiv preprint directly into the ingest pipeline

Do not use this skill when:

• the task is mainly about searching the local knowledge base
• the user wants a full multi-source literature survey; use search or explore

Core Workflow

1. Decide whether this is browse/search or fetch/ingest

• If the user wants to look around, discover preprints, or check what is new, start with arxiv search
• If the user wants cross-source retrieval with the local library, use fsearch --scope main,arxiv
• If the user already has a specific arXiv paper in mind, use arxiv fetch
• If the goal is "put this preprint into my library now", use arxiv fetch <id> --ingest

2. Use the right command

Direct arXiv search:

scholaraio arxiv search "<query>"
scholaraio arxiv search "<query>" --category physics.flu-dyn
scholaraio arxiv search --category cs.LG --sort recent

Federated search when arXiv should complement the local library:

scholaraio fsearch "<query>" --scope main,arxiv
scholaraio fsearch "<query>" --scope arxiv

Fetch PDF only:

scholaraio arxiv fetch 2603.25200
scholaraio arxiv fetch arXiv:2603.25200v1
scholaraio arxiv fetch https://arxiv.org/abs/2603.25200v1

Fetch and ingest immediately:

scholaraio arxiv fetch 2603.25200 --ingest

3. Keep the source semantics clear

• arXiv is a preprint source, not the main curated knowledge base
• Use it to discover fresh work quickly or pull in a specific preprint
• If the paper becomes relevant to the user's ongoing work, ingest it instead of repeatedly treating it as an external result

Practical Heuristics

• If the user says "也搜一下 arXiv", prefer fsearch --scope main,arxiv
• If the user says "看看最近有什么预印本", prefer arxiv search --sort recent
• If the user gives an arXiv ID or URL, do not start with broad search; go straight to arxiv fetch
• If the user is clearly collecting papers for continued reading or writing, prefer --ingest

Output Style

• Make it clear when results come from arXiv rather than the local library
• When using federated search, distinguish local hits from arXiv-only hits
• When fetching without ingest, mention that the PDF now sits in the configured paper inbox (fresh default: data/spool/inbox/)
• When fetching with ingest, mention that the paper has been sent into the normal ingest flow