基于 anchor 论文、topic 关键词或当前 wiki 状态，产出一份排好序的候选论文 shortlist，供用户或上游 skill 决定是否进一步 `/ingest`。当用户问 "接下来该读什么"、"找和这篇相似的论文"、"推荐相关工作"、"这个方向周围有什么" 时触发；`/ingest --discover` 也会内部调用本 skill。本身不 ingest，只提出候选。

To generate a ranked paper reading shortlist, provide anchor paper IDs or topic keywords. The system retrieves semantically similar work and literature context, producing ordered candidate papers with rationales while ensuring nothing is ingested automatically.

Yes, you can find related academic papers without automatic ingestion. The discovery process proposes ranked candidates and logs events to wiki/log.md, strictly avoiding modifications to raw files or automatic wiki content updates.

Discovering next readings based on recently modified wiki pages requires selecting the wiki exploration seed mode. The system derives plausible next reading suggestions from modified wiki paper pages and ranks them with clear rationales.

Wiki deduplication during literature recommendation filters out papers already present in the wiki. It checks for matches using arxiv or arxiv_id frontmatter to ensure candidate shortlists contain only new, unread papers.

Yes, the paper discovery process supports topic keyword searches for literature reviews. You can provide a topic query string to retrieve and rank plausible next readings with clear rationales for your review.

A limitation of using a topic query for paper discovery is that you must select exactly one seed mode. The system cannot mix anchor IDs, topic keywords, and wiki page seeds simultaneously to generate the candidate shortlist.