Imported from 050817/daily-arxiv-briefing-agent (AGENTS.md). Install upstream with npx skills add 050817/daily-arxiv-briefing-agent. Copyright stays with the author.

Daily arXiv Briefing Agent

You are the Daily arXiv Briefing Agent, a three-skill workflow for turning a research query into an evidence-grounded arXiv briefing with paper metadata, relevance ranking, Markdown/PDF reports, keyword graphs, archives, and grounded file Q&A.

Available Skills

• arxiv-paper-retrieval - Retrieve arXiv papers, normalize metadata, apply date filters, and save data/raw/arxiv_papers.json
• paper-relevance-ranking - Rank retrieved papers with a local TF-IDF baseline and save data/processed/ranked_papers.json
• arxiv-briefing-graph - Generate paper cards, report-level insights, Markdown/PDF reports, SVG keyword figures, archives, and archive Q&A

How it works

• The user provides a research query, date range, result limit, and Top-K size
• Skill 1 retrieves candidate arXiv paper metadata, optionally using an OpenAI-compatible model for keyword extraction
• Skill 2 ranks the retrieved papers by query relevance without requiring model or network access
• Skill 3 creates the briefing, figures, archive outputs, and file-aware chat; when configured, it uses an OpenAI-compatible model for summaries, trend interpretation, reading order, limitations, and archive Q&A
• The workflow can run end to end from main.py, or each skill can run independently for stage-level experiments
• Existing archives can be queried directly with Skill 3 Q&A by providing archive_path and message; this does not rerun retrieval, ranking, or report generation
• Model settings can come from web_app/local_settings.json or OPENAI_API_KEY, OPENAI_BASE_URL, and OPENAI_MODEL

Run

python main.py --query "graph neural networks for misinformation detection" --date_range "last 7 days" --max_results 10 --top_k 5

Run an individual skill:

python skills/paper_retrieval/skill.py --query "graph neural networks" --date_range "last 7 days" --max_results 10
python skills/relevance_ranking/skill.py --input data/raw/arxiv_papers.json --query "graph neural networks" --top_k 5
python skills/briefing_graph/skill.py --input data/processed/ranked_papers.json --query "graph neural networks" --top_k 5

Ask a question about an existing archive without rerunning the full workflow:

from skills.briefing_graph.skill import BriefingGraphSkill

skill = BriefingGraphSkill()
result = skill.chat_with_archive(
    "archives/graph-neural-networks_20260515_103000",
    "What are the most relevant papers in this archive?"
)
print(result["answer"])

Output Files

• data/raw/arxiv_papers.json
• data/processed/ranked_papers.json
• outputs/reports/daily_briefing.md
• outputs/reports/daily_briefing.pdf
• outputs/figures/keyword_graph.svg
• outputs/figures/top_keywords.svg
• archives/<query-and-time>/