Turn a short topic into a verified review package

A researcher made an agent skill that produces literature reviews from a brief topic and saves them in several formats.

Done with Custom agent· Advanced· Oct 2026

What they did
They built the skill into an agent framework so a user can start with a single topic line. The process begins by pinning down the scope through four keyword layers, a freshness setting, and a route that directs work to arXiv or PubMed depending on the field. After that, it runs ten scripts in five rounds against OpenAlex, Semantic Scholar, CrossRef, Unpaywall, arXiv, and PubMed, then strips duplicates by matching titles and DOIs while relying on API metadata instead of invented details.
How it went
The author has used and improved it with their team for more than six months, and students have liked the output. Every reference includes either a DOI or a URL. The Feishu version is easier to share, but it does not perform as well as the local one.
Tips
Keep the search logic tied to source data and duplicate filtering, not to model memory. If you want wide sharing, be ready for a web-hosted version to be less capable than a local build.

Try it yourself with Claude Code

Build an agent skill for academic reviewing. Start from [topic], then ask me to lock in four tiers of search terms, a recency setting, and the source path for [field], using [arXiv/PubMed/other]. Pull only real papers from the chosen services, compare titles and DOI/URL fields to drop duplicates, and write the review in [language] with [citation style] formatting. Export the finished work as Markdown, Word, and PDF. Before any irreversible action, including saving over files, sending content, submitting a form, or deleting anything, stop and ask me to approve it first.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on wangshuyi.substack.com ↗

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.