Fetch RSS Feeds

Fetch and store articles from RSS/Atom feeds with deduplication.

When to Use

Use this skill when you need to:

• Collect articles from RSS or Atom feeds
• Store articles in memory for later analysis
• Avoid processing duplicate articles

Instructions

Step 1: Check for Duplicates Before Fetching

Before fetching each feed, prepare to deduplicate articles:

• Use memory/check_seen with the article URL as the key
• Namespace: news/articles

Step 2: Fetch RSS Feeds

Use the rss tool to fetch articles from configured feeds.

Feed sources to fetch:

• Company blogs (OpenAI, Anthropic, Google DeepMind, etc.)
• AI-focused tech news (TechCrunch AI, VentureBeat AI, The Verge AI)
• Research feeds (arXiv, Papers with Code)

For each feed:

1. Call the rss tool with the feed URL
2. Extract: title, url, published_date, summary, author

Step 3: Filter and Deduplicate

For each article from the feed:

1. Check if already seen:

   • Call memory/check_seen with key=URL hash, namespace="news/articles"
   • If seen=true, skip this article

2. Validate required fields:

   • Article must have: title, url
   • Skip articles missing required fields

3. Filter by age (optional):

   • If max_age_hours is specified, skip articles older than threshold

Step 4: Store New Articles

For each new (unseen) article:

1. Store in memory:

   • Call memory/add with:
     • type: "document"
     • namespace: "news/articles"
     • data: {title, url, summary, author, source, published_at}
     • metadata: {fetched_at, source_feed}

2. Mark as seen:

   • Call memory/mark_seen with:
     • key: URL hash
     • namespace: "news/articles"
     • ttl_seconds: 1209600 (14 days)

Step 5: Return Results

Return a summary including:

• Number of articles stored
• Number of duplicates skipped
• Number of feeds processed
• Any failed feeds

Tool Usage Guidance

rss tool

• Use to fetch feed content
• Handles XML parsing and date extraction
• Returns list of entries with title, link, summary, published

memory/check_seen

• Call before processing each article
• Key should be a hash or the URL itself
• Returns {seen: true/false}

memory/add

• Store each new article
• Include all extracted metadata
• Type should be "document"

memory/mark_seen

• Call after successfully storing
• Use 14-day TTL to allow re-processing after expiry

Error Handling

• If a feed fails to fetch, log the error and continue with other feeds
• If memory operations fail, log but don't crash
• Return partial results if some feeds succeed

Success Criteria

• At least one feed successfully fetched
• New articles stored in memory
• Duplicates correctly identified and skipped
• Failed feeds logged but don't stop collection