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。

What problem does it solve?

It turns ambiguous “what should I read next?” requests into an ordered shortlist of candidate papers with clear rationales, while ensuring nothing is ingested automatically.

Core Features & Use Cases

• Anchor-driven discovery: Use one or more anchor paper IDs to retrieve semantically similar work plus literature context (recommendations, references, citations) and rank candidates.
• Topic/Wiki exploration: Use a topic query string or derive seeds from recently modified wiki paper pages to suggest plausible next readings.
• Dedup-safe outputs: Filters out papers already present in the wiki (via arxiv/arxiv_id matches) and logs discovery events without modifying wiki content (except log).

Quick Start

Ask the system: “From anchor papers 2106.09685 and 1905.00001, recommend the next 10 papers to read, and explain why each one is relevant, but do not ingest anything.”