A research assistant that uses Firecrawl to monitor the latest publications in PNAS (Proceedings of the National Academy of Sciences) and summarize the most relevant papers based on your background in plant genomics, DNA language models, and AI.
This skill configures the agent to automatically scrape the recent publications from PNAS and filter them based on your specific research expertise in plant genetics, DNA language models, and machine learning.
docker compose up -d) and configured in the mcp_config.json.When the user invokes this skill, follow these exact steps:
Scrape Journal Pages: Use the tool to fetch articles from of the following PNAS pages:
firecrawl_scrapea. Current print issue TOC (weekly articles):
firecrawl_scrape(url="https://www.pnas.org/toc/pnas/current", formats=["markdown"])
b. Latest Articles (articles published online ahead of print — these never appear in the weekly TOC until their print date):
firecrawl_scrape(url="https://www.pnas.org/action/showFeed?type=etoc&feed=rss&jc=pnas", formats=["markdown"])
Or alternatively scrape: https://www.pnas.org/latest-articles
Combine the article lists from both sources before filtering. If either URL is blocked or returns insufficient content, fall back to firecrawl_search with a query like "site:pnas.org research article 2026".
Filter & Select: Read the scraped content and identify all relevant papers based on the user's core research background. The core research interests to prioritize are:
firecrawl_search, specifically query for the "latest issue" or current month/year to avoid retrieving older highly-cited papers.Fetch Abstracts:
For each relevant paper, use the firecrawl_scrape tool to fetch its dedicated abstract page on the PNAS website to get the full abstract text and DOI.
3b. Deduplicate Against Existing Papers:
Before presenting results, check the existing database at docs/js/papers.json in the repository. Read the file and extract all existing paper DOIs and titles. Cross-reference your candidate papers against this list — if a paper's DOI or title (case-insensitive) already exists in papers.json, skip it and do not include it in the final results. Only present papers that are genuinely new and not already tracked.
Format Output: Present the final output using clear markdown headings and bullet points. For each paper, you MUST include:
### 1. [Paper Title](https://www.pnas.org/doi/XXXX) - *[Journal Name]*
* **Authors:** Author 1, Author 2, etc.
* **Abstract Summary:** [2-3 sentences summarising the abstract and main findings]
* **Relevance:** [1 sentence explaining why it aligns with plant genomics or biological AI]
* **Link:** [Abstract Page](https://www.pnas.org/doi/XXXX)
After presenting the results, always ask:
"Would you like to add any of these to your paper notebook website? Reply with the numbers (e.g. 1, 3), all, or none."
Then follow the Add to Notebook skill (.agents/skills/add_to_notebook/SKILL.md) to handle selection, schema building, and automatic commit + push to GitHub.