Arxiv Scout

arxiv-scout

Purpose

Monitors arXiv for newly submitted papers tagged with agentic AI or MCP keywords. Appends deduplicated markdown notes to a local knowledge-base file.

Runbook

1. Verify that curl, jq, and python3 are available; abort if any are missing.
2. Confirm MEMORY_PATH points to a writable file (create it empty if absent).
3. Run scripts/fetch.sh to query the arXiv Atom API for papers matching: ti:agentic OR ti:MCP OR abs:model-context-protocol OR abs:agentic-AI with max_results=20 ordered by submittedDate descending.
4. Inspect the returned JSON: each entry should have id, title, authors, published, summary, and link.
5. Run scripts/sync-notes.sh to deduplicate against existing entries in MEMORY_PATH by arxiv ID, then append new formatted notes.
6. Print a summary: N new papers appended, M duplicates skipped.
7. Optionally, if TELEGRAM_BOT_TOKEN and TELEGRAM_CHAT_ID are set, send the summary line to Telegram via curl.

Stop conditions

1. Abort if curl or jq is missing from PATH.
2. Abort if MEMORY_PATH is unset or resolves to a directory.
3. Abort if the arXiv API returns HTTP status other than 200 (unless MOCK_MODE=1).
4. Never overwrite MEMORY_PATH — only append; abort if the file is not writable.

Output format

{
  "fetched": 20,
  "new": 5,
  "skipped": 15,
  "appended_ids": ["2406.12345", "2406.12346"],
  "memory_path": "/path/to/MEMORY.md"
}

A markdown note block per new paper is appended to MEMORY_PATH:

## [2406.12345] Agentic AI for Code Review
**Authors**: Alice Smith, Bob Jones  
**Published**: 2024-06-15  
**Summary**: We present a multi-agent framework for automated code review...  
**Link**: https://arxiv.org/abs/2406.12345

Example invocations

• MEMORY_PATH=~/notes/MEMORY.md skills/arxiv-scout/scripts/run.sh
• MOCK_MODE=1 skills/arxiv-scout/scripts/fetch.sh
• "Run arxiv-scout and tell me what new MCP papers were published this week."
• "Check arXiv for agentic AI papers and append summaries to my knowledge base."