🇻🇳 Tiếng Việt

← Home · Installation · Architecture · First Agent · Google · Browser · Cron · Multi-Agent · Profiles · Models · Memory · Ops

Recipe: Content Digest (Reddit + YouTube)

Automated daily digest of Reddit posts and YouTube videos, translated and summarized.

Architecture

Cron trigger → Agent runs script → Script fetches content →
Agent translates/summarizes → Upload to Google Drive → Send Telegram summary

Reddit Digest

Script Structure

~/.openclaw/common-scripts/reddit/
├── digest-workflow.sh           # Orchestrator
├── generate-detailed-digest.js  # Fetch + format to markdown
└── (node_modules via shared .venv)

Cron Job

{
  "id": "reddit-digest",
  "agentId": "personal",
  "schedule": { "kind": "cron", "expr": "0 10 * * *", "tz": "Your/Timezone" },
  "payload": {
    "kind": "agentTurn",
    "message": "Run Reddit digest:\n1. Execute: bash ~/.openclaw/common-scripts/reddit/digest-workflow.sh ~/.openclaw/workspace-personal reddit\n2. Read the generated markdown\n3. Translate headlines and summarize each post\n4. Re-upload to Google Drive\n5. Send Telegram with Doc link + top trends",
    "timeoutSeconds": 900
  },
  "delivery": { "mode": "announce", "channel": "telegram", "to": "YOUR_ID", "bestEffort": true }
}

Rate Limiting

• Reddit API: Use reddit-readonly skill (no auth needed for public content)
• Don’t scrape more than 5 posts × 10 comments per subreddit per run

YouTube Digest

Transcript Pipeline

yt-dlp (list videos) → youtube-transcript-api (subtitles) → MLX Whisper (fallback)

• Free transcripts via youtube-transcript-api (no API key)
• Fallback: MLX Whisper runs locally on Apple Silicon (~30-90s per video)
• Rate limit: 5s sleep between transcript requests, max 24/session

Tips

• Monitor channel via RSS — cheaper than API
• Cache seen video IDs in memory/seen-videos.txt to avoid re-processing
• Long videos (>30min): Use transcript, don’t try to summarize from title alone