arxiv-to-zotero（arxiv 到 zotero）

arXiv to Zotero

Use this skill when the user wants recent arXiv papers found and written into Zotero directly.

Best-fit use cases

• find papers by topic + time range
• skip papers already in Zotero
• attach PDFs when possible
• keep imports organized with a fixed tag and a dedicated collection

Setup

On first use, if ~/.openclaw/config/skills/arxiv-to-zotero.setup.json is missing, read {baseDir}/setup.md, collect the required Zotero values, tell the user to put ZOTERO_API_KEY in ~/.openclaw/.env, create the setup-state file, and resume the original request exactly once.

Flow

1. If the user has not already provided them, ask for:

- topic keywords or phrases - a time range

1. Do not ask the user to write an arXiv query.
2. If the user gives keywords in Chinese, translate them into concise, technically accurate English search phrases before building the arXiv query.
3. Build one valid arXiv API search_query yourself.
4. Run the script once:

python3 {baseDir}/scripts/main.py --config {baseDir}/config.json --query '<arXiv search_query>'

1. After the script finishes, use result.user_message as the final user-facing notification.

Query rules

• Use official arXiv fielded search syntax.
• The final arXiv query must be written in English.
• Use double quotes for multi-word phrases.
• Combine alternative keywords with OR when appropriate.
• Add the requested time range with submittedDate:[YYYYMMDDTTTT TO YYYYMMDDTTTT].
• Pass the query as normal text. Do not URL-encode it yourself.

Guarantees

• Existing Zotero items are not modified.
• New parent items are tagged with arxiv-to-zotero by default.
• New imports are placed into the arxiv-to-zotero collection by default. If that collection does not exist, the script creates it.
• If Zotero file upload hits HTTP 413, the script falls back to linked_url for that paper and keeps later attachments in link mode for the same run.

When not to use

Do not use this skill for discussion-only requests, browsing help, or any task that should not write to Zotero.

Network / Privacy

• Contacts: arXiv Atom API, arXiv PDF URLs, Zotero Web API
• Secret used: ZOTERO_API_KEY
• Writes: new Zotero parent items and child attachments only
• Does not modify existing Zotero items

Natural-language trigger examples

• 帮我找近三年来 mamba 或者多模态用于股票预测的 arXiv 论文，并导入 Zotero。
• 帮我查最近两年 test-time adaptation 或 active search 用于组合优化的 arXiv 论文，去重后导入 Zotero。
• 帮我找近半年的 graph neural network 用于 TSP 或 vehicle routing 的 arXiv 论文，导入 Zotero。