Explore

You are discovering new RSS/Atom feed sources relevant to the user's interests. Follow these steps exactly:

Step 1: Understand what's already covered

Read sources.json and topics.json from the project root.

From sources.json, build a list of existing source names and categories. From topics.json, understand the user's roles and topic priorities.

Identify coverage gaps — topic areas from topics.json that have few or no sources in sources.json. For example, if "AI coding tools" is high priority but only one AI source exists, that's a gap.

Step 2: Research new sources

Use WebSearch to find high-quality blogs, newsletters, and technical publications that:

• Cover the user's high and medium priority topics
• Fill identified coverage gaps
• Are written at a senior engineering level (not beginner tutorials)
• Have active RSS/Atom feeds
• Are actively maintained (published within the last 3 months)

Search strategies (run these in parallel where possible):

• "best RSS feeds" + {topic} for each gap area
• "engineering blog" RSS + {topic}
• "{topic} newsletter" RSS feed
• Check well-known aggregators (e.g., bloggingfordevs.com, feedle.world) for curated lists

Aim to find 5-10 candidate sources that aren't already in sources.json.

Step 3: Validate each candidate

For each candidate source, perform ALL of these validation checks:

3a. Feed URL validation

Use Bash with curl to test the feed URL:

curl -sL --max-time 10 -o /dev/null -w "%{http_code}" "{feed_url}"

The feed must return HTTP 200. If not, try common alternatives (/feed, /feed.xml, /rss, /rss.xml, /atom.xml, /index.xml).

3b. Feed content validation

Fetch the first 30 lines of the feed and verify it's valid XML/RSS/Atom:

curl -sL --max-time 10 "{feed_url}" | head -30

Must contain <rss, <feed, or <channel tags. Skip feeds that return HTML, JSON, or error pages.

3c. Freshness check

Use WebFetch on the feed URL to check when the most recent entry was published. Skip sources with no posts in the last 3 months.

3d. Content quality check

Use WebFetch to read the source's main page or a recent article. Verify:

• Content is senior-level (not "Learn JavaScript in 10 days")
• Posts have technical depth (not just link aggregation or product announcements)
• Not primarily a sales/marketing blog

Run feed URL and content validations in parallel across candidates (use direct parallel tool calls, NOT the Agent tool).

Step 4: Present findings

Output a summary table to the user:

## Source Discovery Results

### Validated Sources (ready to add)

| # | Name | Category | Type | Feed URL | Why |
|---|------|----------|------|----------|-----|
| 1 | ... | ... | article/newsletter | ... | ... |

### Rejected Candidates

| Name | Reason |
|------|--------|
| ... | Feed returns 404 / Stale (last post 6 months ago) / Beginner content / etc. |

For each validated source, include:

• Name: The blog/newsletter name
• Category: Best matching category from existing sources.json categories, or suggest a new one
• Type: article or newsletter (set newsletter if it's a curated link roundup)
• Full content RSS: Whether the feed includes full article text (set full_content_rss: true if so)
• Feed URL: The validated URL
• Why: 1 sentence on what gap it fills

Step 5: Add approved sources

Ask the user which sources (by number) they want to add. Accept a response like "1, 3, 5" or "all".

Then for each approved source, add it to sources.json with the correct structure:

{
  "name": "Source Name",
  "url": "https://example.com/feed.xml",
  "category": "Category",
  "enabled": true
}

Include "type": "newsletter" if applicable. Include "full_content_rss": true if applicable. Only add fields that are needed.

After updating sources.json, run a quick test:

cd $PROJECT_ROOT && python3 -c "
import json
with open('sources.json') as f:
    data = json.load(f)
enabled = [s for s in data['sources'] if s.get('enabled', True)]
print(f'{len(enabled)} enabled sources, {len(data[\"sources\"])} total')
"

Confirm to the user how many sources were added and the new total.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: rtivital
• Source: rtivital/claude-rss-digest-skill
• License: MIT

Install and usage instructions live in the source repository linked above.