Summarize new arXiv papers into notes with scheduled agents
Scheduled scripts produce a daily arXiv digest, and a nightly script launches Claude Code agents to summarize each saved PDF into a notes folder.
What they did
A Python script driven by a topic file searches arXiv each morning, and Google Scholar weekly. It skips papers it has already seen and writes a dated Markdown digest. The user scans the digest and saves worthwhile PDFs into topic source folders. A nightly script then starts a Claude Code agent for each new PDF, using a summary skill, and saves the summary in the matching notes folder.
How it went
The source gives no measured results. It describes the target as each saved PDF being summarized by the next digest run, with summaries covering method, sample, findings, effect sizes and limitations.
Worth knowing
Keep the manual PDF selection step. It keeps the summarization queue small, and a vague summary prompt tends to just restate the abstract.
Labels
A nightly script then starts a Claude Code agent for each new PDF, using a summary skill, and saves the summary in the matching notes folder.
a nightly script launches Claude Code agents to summarize each saved PDF into a notes folder.
The “Material” label failed its quality check and is not used for selection.
Details
- Agents
- Claude Code
- Source
- other
- Source date
- undated
- Accounts
- 1
- In registry since
- Oct 12, 2026