Ground literature work in paper indexes and paper text.
Search alphaXiv and OpenAlex together — full-text and citation-graph hits cover different misses.
Read the alphaXiv overview to orient, then local arxiv2md for evidence, because overview text is not the paper.
Do not start with ad-hoc web search or raw PDF/HTML fetches from arxiv.org.
After the text is in hand, use research-taste for critique and taste.
For .bib / \cite{...}, use bibcite — this skill does not edit bibliographies.
Search
Run both in parallel for a topic or method phrase:
curl -sG "https://api.alphaxiv.org/search/v2/paper/full-text" \
--data-urlencode "q=<terms>" \
--data-urlencode "limit=10"
curl -s "https://api.openalex.org/works?filter=title_and_abstract.search:<terms>&per_page=20&select=id,title,publication_year,cited_by_count,ids&api_key=$OPENALEX_API_KEY"
Omit &api_key=... when OPENALEX_API_KEY is unset.
URL-encode query terms.
Use alphaXiv when the query is wording that lives in the body; use OpenAlex when you need lineage, citation weight, or a seed to expand.
Merge by versionless arXiv id; keep OpenAlex-only works that have no arXiv id.
Return a short ranked list (id, title, year, why it matched) — not raw JSON.
Delegate multi-paper sweeps to subagents so the main context keeps findings.
Read references/search.md when walking a seed's citation graph, decoding OpenAlex abstracts, handling slug/non-arXiv alphaXiv ids, or both indexes miss.
Read
Normalize any arXiv / alphaXiv URL to an id (last path segment; strip .pdf / .md).
Classic ids look like 2401.12345 or 2401.12345v2.
Overview first — cheap map of the paper:
curl -sL "https://www.alphaxiv.org/overview/<id>.md"
Then arxiv2md for the sections the question needs — so context stays on evidence, not the whole PDF:
uvx --from arxiv2markdown arxiv2md <id> \
--section-filter-mode include \
--sections "Abstract,Introduction,Experiments" \
-o paper.md
If section titles are unknown, convert with --remove-refs --remove-toc, then narrow.
Do not call https://arxiv2md.org/api/....
Read paper.md.
Numbers, equations, tables, hyperparameters, and experimental setup must come from arxiv2md (or a figure you opened), not from the overview alone.
If overview 404s on a classic id, go straight to arxiv2md.
Read references/reading.md for figures, slug ids, linked code, or when arxiv2md has no HTML.