Search arXiv and Trace Academic Citations with Hermes

The bundled arXiv skill combines the arXiv Atom API with Semantic Scholar lookups. It supports fielded and Boolean search, sorting, pagination, exact paper retrieval, BibTeX generation, citation and reference graphs, recommendations, and author profiles.

Official skill reference: arXiv Research

Build a reproducible search

Use title, author, category, or all-field queries deliberately. Record the query, sort order, date, and result limit. Retrieve promising papers by stable arXiv ID, then read the abstract and full paper before using the work as evidence.

Semantic Scholar can add citation counts, references, citing papers, related work, and author information. Those signals help discovery, but a highly cited paper is not automatically correct or relevant. Check version suffixes and withdrawal notices; an arXiv identifier can point to revised versions over time.

Research checklist

• Write the research question before searching.
• Record exact query syntax, categories, sort, and retrieval date.
• Deduplicate papers across arXiv IDs, DOIs, and title variants.
• Verify title, authors, year, and identifier from live metadata.
• Read the claim in the paper before citing it.
• Trace both references and later citations when context matters.
• Export BibTeX from verified metadata rather than inventing it.

Common pitfalls

• Searching only broad keywords. Fielded queries reduce unrelated results.
• Citing the abstract for a detailed methodological claim. Read the relevant section of the paper.
• Equating citation count with quality. Counts are field- and age-dependent and can include criticism.
• Ignoring revised or withdrawn versions. Record the exact version used.
• Hammering public APIs. Respect documented rate limits and cache stable metadata.

Verification

Open each final paper by identifier, compare metadata across arXiv and a second source when available, and verify every cited claim against the paper text. Run a duplicate check on the bibliography and preserve search details so another researcher can repeat the discovery path.