Imported from oimiragieo/agent-studio (.claude/skills/exa-monitor/SKILL.md). Install upstream with npx skills add oimiragieo/agent-studio --skill exa-monitor. Copyright stays with the author.

Exa Monitor Skill

Polls the Exa search engine every 4 hours for topics configured in EXA_MONITOR_TOPICS. Deduplicates results against previously seen URLs, stores new summaries in named memory, and integrates with the morning briefing loop.

Setup

1. Set EXA_MONITOR_TOPICS in .env:

   EXA_MONITOR_TOPICS=["Claude AI updates","agent frameworks","LLM tooling","AI safety"]
   

2. Start the monitor loop:

   /loop 4h Skill({ skill: 'exa-monitor' })
   

   Or via CronCreate:

   CronCreate({
     schedule: '0 */4 * * *',
     task: "Invoke Skill({ skill: 'exa-monitor' }) to fetch new Exa search results",
   });

Core Logic

Step 1: Load Topics and Seen URLs

let topics;
try {
  topics = JSON.parse(process.env.EXA_MONITOR_TOPICS || '["Claude AI updates","agent frameworks"]');
} catch (_e) {
  topics = ['Claude AI updates', 'agent frameworks'];
}

// Load previously seen URLs from memory
const seenRaw = await readMemory('exa-seen-urls');
const seenUrls = new Set(seenRaw ? JSON.parse(seenRaw) : []);

Step 2: Search Each Topic via Exa MCP

// Use Exa MCP tool (preferred — returns structured results)
// Skill({ skill: 'exa-monitor' }) invokes mcp__Exa__web_search_exa internally:
// mcp__Exa__web_search_exa({ query: topic, numResults: 5, useAutoprompt: true })

// Falls back to mcp__Exa__get_code_context_exa for technical topics

Step 3: Filter and Deduplicate

const newResults = [];
for (const result of exaResults) {
  if (seenUrls.has(result.url)) continue;
  seenUrls.add(result.url);
  newResults.push({
    title: result.title,
    url: result.url,
    summary: result.text?.slice(0, 400) || result.highlights?.join(' ') || '',
    topic,
    publishedDate: result.publishedDate,
  });
}

// Persist seen URLs via named memory (cap at 2000)
const seenArr = [...seenUrls].slice(-2000);
await writeMemory('exa-seen-urls', JSON.stringify(seenArr));

Step 4: Append to Digest

if (newResults.length > 0) {
  const digest = newResults
    .map(
      r =>
        `## ${r.title}\n**Topic:** ${r.topic}\n**Published:** ${r.publishedDate || 'unknown'}\n${r.summary}...\n[Read →](${r.url})\n`
    )
    .join('\n---\n');

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

Integration with Morning Briefing

/loop at 8:00am Read .claude/context/memory/named/exa-digest.md and arxiv-digest.md. Summarize the most relevant news and papers for agent-studio development. Highlight any urgent developments.

Configuration Reference

Deduplication

• Seen URLs stored via writeMemory('exa-seen-urls', ...) (named memory API)
• Capped at 2000 most recent URLs
• Reset with: await writeMemory('exa-seen-urls', '[]')

Related Skills

• scheduled-tasks — CronCreate API
• arxiv-monitor — ArXiv companion monitor for academic papers
• heartbeat — Full heartbeat ecosystem including this loop
• memory-search — Search the exa-digest for specific topics