dsh-arxiv-search

arXiv search tool plugin for the DeepSeek Harness (DSH). Registers one model-facing tool, arxiv_search, that queries the public arXiv API and returns structured paper metadata — no API key, no Python dependency, no client bundle.

Originally adapted from the arxiv-search skill (MIT, aiskillstore marketplace).

Install

dsh plugin --profile web add https://github.com/<your-account>/dsh-arxiv-search/archive/refs/tags/v0.1.0.tar.gz

Then restart the web server (dsh web) so the Host loads the new plugin. In any new conversation the model can call the arxiv_search tool directly.

Usage

Just ask in natural language, e.g.:

• "搜一下 2024 年后的 uplift modeling 论文"
• "Find recent papers about diffusion models, 5 results"
• "按提交时间排序搜 protein folding"

Tool arguments:

Arg	Type	Notes
query	string (required)	Search query, e.g. "uplift modeling"
maxPapers	number	Max results (default 10, capped 200)
sortBy	string	relevance | submittedDate | updatedDate (default relevance)

Each result includes title, authors, abstract, dates, categories, abs URL, and PDF link.

Uninstall

Remove the dependency from $DSH_HOME/profiles/<profile>/package.json (and from dsh.profile.bundles), run pnpm install in that directory, then restart the web server.

Development

Plain ESM JavaScript, no build step — lib/ is the source of truth and ships as-is. To test the search logic standalone:

node --input-type=module -e "import { searchArxiv } from './lib/index.js'; console.log(await searchArxiv({ query: 'uplift modeling', maxPapers: 3 }))"

License

MIT. The adapted skill content originates from the aiskillstore marketplace (MIT).