orx-lit-review
Search and read research papers. The main agent calls alphaXiv, OpenAlex, and bioRxiv discovery primitives, ranks the combined candidates, and chooses sources for focused follow-ups. Use for literature reviews, related work, prior art, papers, authors, methods, benchmarks, or research claims; never delegate the retrieval loop to a sub-agent.
alphaXiv · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add alphaXiv/OpenResearch --skill orx-lit-review技能说明
name: orx-lit-review description: "Search and read research papers. The main agent calls alphaXiv, OpenAlex, and bioRxiv discovery primitives, ranks the combined candidates, and chooses sources for focused follow-ups. Use for literature reviews, related work, prior art, papers, authors, methods, benchmarks, or research claims; never delegate the retrieval loop to a sub-agent."
Literature retrieval
You are the retrieval ranker. Call the alphaXiv, OpenAlex, and bioRxiv primitives yourself, inspect the returned candidates, and decide which sources are useful for each focused follow-up. Never delegate this loop to a sub-agent.
Each command performs exactly one public endpoint request and emits its structured JSON result. No login is required:
orx discover keyword "<exact keyword query>"
orx discover embedding "<semantic description in the user's terms>"
orx discover openalex "<scholarly search query>"
orx discover biorxiv "<biology preprint query>"
keywordsearches title, abstract, and full text. Results include the match snippets that explain why each paper was retrieved. Use short exact terms: method names, acronyms, benchmarks, authors, or title phrases. Use only terms stated by the user or observed in results; never invent an acronym expansion.embeddingsearches titles and abstracts semantically, then reranks by similarity and the requested priority. Use the user's actual question or a concise description of a genuinely missing facet.openalexsearches the cross-disciplinary OpenAlex scholarly graph. It is especially useful for journal/conference papers, citation context, and work outside arXiv.biorxivsearches OpenAlex's bioRxiv source index. bioRxiv has no comparable native search API; the bioRxiv API is used later when reading a selected DOI.- Every primitive returns the same JSON shape:
source, self-routingid, title, abstract, and publication date. alphaXiv results may include votes and full-text snippets; OpenAlex and bioRxiv results may include citations.
Date and ranking controls
Retrieval is not date-bounded unless you supply a bound. Add the same controls to any primitive when the question calls for them:
orx discover keyword "<query>" --published-after 2024-01-01 --prioritize recency
orx discover embedding "<query>" --published-before 2012-01-01 --prioritize historical
orx discover openalex "<query>" --published-after 2024-01-01 --prioritize recency
orx discover biorxiv "<query>" --limit 20
--published-afterand--published-beforeare inclusiveYYYY-MM-DDbounds. Do not invent a cutoff merely to favour newer work.- Older or narrow
--published-beforeembedding searches can return a thin or empty candidate set because the upper bound is applied after vector retrieval. Report what comes back; do not treat an empty set as proof that no literature exists or retry the identical query and window. --prioritizeisdefault,recency,historical, orpopular.--limitcan narrow alphaXiv output but cannot widen alphaXiv's fixed server-side candidate pools. For OpenAlex and bioRxiv it also controls the requested pool size.- OpenAlex and bioRxiv implement these controls with OpenAlex publication-date filters, then rerank the returned relevance pool by date or citations. Their ranking is best-effort and is not identical to alphaXiv's semantic, vote-aware ranking.
- Use
recencyfor explicitly new/latest work. Usehistoricalfor seminal or foundational work. Usepopularonly when the user asks about votes, popularity, or community standing.
Main-agent retrieval loop
You are the low-latency retrieval ranker. Run the loop below yourself.
Set up the retrieval query
- If using keyword retrieval, build focused terms using only wording from the user or prior tool results. Never guess an acronym expansion. General-purpose padding reduces result quality.
- For semantic or scholarly-graph retrieval, build one short faithful question in the user's terms rather than a padded reformulation.
- Estimate retrieval difficulty from 1–10. This controls a budget of complete follow-up rounds: difficulty 1–3 gets 0 rounds, 4–7 gets 1, and 8–10 gets 2.
- Resolve one publication window and priority for the request. Every initial and follow-up call must inherit those exact controls; never widen a window or change priority during the loop. Every returned candidate already satisfies that window, so rank what is available instead of lamenting well-known work that the user excluded.
Run and rank
- Choose the initial sources and strategies that fit the query. For arXiv-heavy ML, CS, math, or physics questions, use alphaXiv keyword, embedding, or both according to whether exact full-text evidence, semantic coverage, or both are useful. Add OpenAlex for broader journal, conference, citation, or cross-disciplinary coverage. Use bioRxiv for biology and adjacent life-science preprints, not as a ritual call for unrelated topics. When the corpus is genuinely ambiguous or interdisciplinary, query multiple relevant sources concurrently. If the initial round includes alphaXiv keyword and its terms mix other terms with one or more 2–10 character tokens that start with a letter, contain only letters, digits, or hyphens, and have at least two uppercase letters, concurrently run one additional keyword call whose query is exactly those acronym tokens joined by spaces and nothing else. This recovery call is part of the initial round.
- Treat initial calls independently: retain every successful result set when another call fails. If none returns results and follow-up budget remains, use a round only when a focused recovery query is likely to work.
- Inspect and deduplicate every candidate. Match exact
idfirst, then a DOI or arXiv id visible in the metadata, then exact normalized title as a cross-source fa