oimiragieo/arxiv-monitor

<!-- Agent: nodejs-pro | Task: #12 | Session: 2026-03-08 -->

ArXiv Monitor Skill

Polls the ArXiv API every 6 hours for papers matching configured keywords. Deduplicates against previously seen papers, stores new summaries in named memory, and integrates with the morning briefing loop.

---

Setup

1. Set ARXIV_KEYWORDS in .env:

   ARXIV_KEYWORDS=multi-agent systems,LLM reasoning,autonomous agents,RAG,tool use
   

2. Start the monitor loop:

   /loop 6h Skill({ skill: 'arxiv-monitor' })
   

   Or via CronCreate for programmatic control:

   CronCreate({
     schedule: '0 */6 * * *',
     task: "Invoke Skill({ skill: 'arxiv-monitor' }) to fetch new ArXiv papers",
   });
   

---

Core Logic

Step 1: Load Keywords and Seen Papers

const keywords = (process.env.ARXIV_KEYWORDS || 'multi-agent systems,autonomous agents')
  .split(',')
  .map(k => k.trim());

// Load previously seen paper IDs from memory
const seenRaw = await readMemory('arxiv-seen-ids');
const seenIds = new Set(seenRaw ? JSON.parse(seenRaw) : []);

Step 2: Search ArXiv API for Each Keyword

Use Bash to query the ArXiv API (no authentication required):

# Search for papers from last 7 days matching a keyword
ENCODED=$(python3 -c "import urllib.parse,sys; print(urllib.parse.quote(sys.argv[1]))" "$KEYWORD")
curl -s "https://export.arxiv.org/api/query?search_query=all:${ENCODED}&sortBy=submittedDate&sortOrder=descending&max_results=10"

Or use the HuggingFace MCP paper_search tool if available:

// Via mcp__claude_ai_Hugging_Face__paper_search if available
// Falls back to curl-based ArXiv API otherwise

Step 3: Filter and Store New Papers

const newPapers = [];
for (const paper of fetchedPapers) {
  if (seenIds.has(paper.id)) continue; // Skip already processed
  seenIds.add(paper.id);
  newPapers.push({
    id: paper.id,
    title: paper.title,
    authors: paper.authors.slice(0, 3).join(', '),
    summary: paper.summary.slice(0, 300),
    published: paper.published,
    url: paper.url,
    keyword: keyword,
  });
}

// Persist seen IDs via named memory (cap at 1000 to prevent unbounded growth)
const seenArr = [...seenIds].slice(-1000);
await writeMemory('arxiv-seen-ids', JSON.stringify(seenArr));

Step 4: Append to Digest File

if (newPapers.length > 0) {
  const digest = newPapers
    .map(
      p =>
        `## ${p.title}\n**Authors:** ${p.authors}\n**Keyword:** ${p.keyword}\n**Published:** ${p.published}\n${p.summary}...\n[Read →](${p.url})\n`
    )
    .join('\n---\n');

  // Append to named memory digest
  const digestPath = '.claude/context/memory/named/arxiv-digest.md';
  const existing = fs.existsSync(digestPath) ? fs.readFileSync(digestPath, 'utf8') : '';
  fs.writeFileSync(
    '.claude/context/memory/named/arxiv-digest.md',
    `${existing}\n\n## ArXiv Update — ${new Date().toISOString().slice(0, 10)}\n\n${digest}`
  );
}

---

Integration with Morning Briefing

The morning briefing loop reads arxiv-digest.md for recent papers:

/loop at 8:00am Read .claude/context/memory/named/arxiv-digest.md and issues.md. Summarize: (1) top 3 new papers relevant to agent-studio, (2) technical debt, (3) top 2 tasks for today.

---

Configuration Reference

| Variable | Default | Description | | --------------------- | --------------------- | -------------------------------- | | ARXIV_KEYWORDS | multi-agent systems | Comma-separated search keywords | | ARXIV_MAX_RESULTS | 10 | Max papers per keyword per run | | ARXIV_LOOKBACK_DAYS | 7 | Days to look back for new papers |

---

Deduplication

• Seen paper IDs stored via writeMemory('arxiv-seen-ids', ...) (named memory API)
• Capped at 1000 most recent IDs to prevent unbounded growth
• Papers with matching IDs are silently skipped on subsequent runs
• Reset by calling await writeMemory('arxiv-seen-ids', '[]')

---

Related Skills

• scheduled-tasks — CronCreate API for scheduling loops
• exa-monitor — Exa web search companion monitor
• heartbeat — Start all 7 heartbeat loops including this one
• memory-search — Search the arxiv-digest for specific papers