ArXiv Papers
Use the arXiv API (and optional PDF fetch) to locate papers and summarize abstracts for the user. Treat results as preprints—not necessarily peer-reviewed or final.
When to use
- “What’s new on arXiv about …”, “Summarize arXiv:XXXX”, category browsing (e.g.
cs.AI). - Quick orientation before deeper reading—not a substitute for reading the full paper in serious research.
Limits (say explicitly when relevant)
- Coverage: arXiv only; many venues are not there.
- Quality: preprint ≠ endorsed truth; contradictory claims exist.
- Rate / ToS: respect arXiv’s API guidelines; don’t hammer endpoints.
Workflow
-
Run
scripts/search_arxiv.sh "<query>"and parse the returned XML (<entry>,<title>,<summary>, PDF<link>). -
Present title, authors, id, abstract summary, and link to abstract/PDF.
-
If the user wants depth, PDF may be fetched selectively—large files and parsing limits apply.
-
Optionally append notable papers to
memory/RESEARCH_LOG.md(if your environment uses it):### [YYYY-MM-DD] TITLE - **Authors**: … - **Link**: … - **Summary**: …
Examples
- Latest LLM reasoning papers on arXiv.
- “What is paper
2512.08769about?”
Resources
scripts/search_arxiv.sh— thin wrapper over the arXiv API.