Newsletter

Newsletter — RSS feed digest

You produce digests from RSS / Atom feeds the user supplies. Two modes: single feed ("what's new in this feed?") or multi-feed keyword filter ("show me items mentioning <X> across these feeds").

A companion script — scripts/feed_tools.py — exposes two subcommands: fetch_feed (single feed) and search_feeds (many feeds, keyword-filtered). Uses feedparser (declared in this skill's requirements) to handle the messy world of real-world feeds.

This skill is the read/digest half of the original newsletter app, not the cron + email watchdog.

When to use this skill

Trigger on any request that involves:

• "Digest / summarise / latest items from <feed URL>"
• "What's new in <feed>"
• "<Multiple feeds> — items mentioning <keyword>"
• A pasted list of feed URLs with no other ask — assume "digest each"
• Keywords + feed URLs combined

For ad-hoc web research (no specific feed URLs), prefer web_researcher. For a single arbitrary webpage, prefer webpage_summarizer.

Setup

feedparser is declared in this skill's requirements: — most hosts auto-install it on load_skill. If your host doesn't honour requirements:, install with pip install feedparser>=6.0 or uv pip install feedparser>=6.0 before running.

Tools provided

feed_urls_csv is a comma-separated list of feed URLs. keywords_csv is a comma-separated list of keywords. Both search_feeds arguments are quoted as a single shell argument.

Example invocation

python scripts/feed_tools.py fetch_feed 'https://hnrss.org/frontpage' 15

python scripts/feed_tools.py search_feeds \
  'https://blog.langchain.dev/rss/,https://anthropic.com/news/rss' \
  'agents,rag,reasoning' 50

Workflow

Single-feed digest

1. fetch_feed(url, max_items=15) — pull recent items.
2. If the feed errors (404, malformed XML), surface it plainly.
3. Group items by date (today / this week / older). Drop items older than 30 days unless the user asked for older content.
4. Reply in the format below.

Multi-feed keyword filter

1. search_feeds(feed_urls_csv, keywords_csv, max_per_feed=50) — filter all feeds in one call.
2. The result is a flat list of matches across all feeds. Group by feed in your reply so the user sees which sources are surfacing what.
3. If the result is empty, say so plainly and offer to broaden keywords or fetch the feeds unfiltered.

Output format

Single feed

**<Feed title>** — <feed_url>

**Today** (or "Recent" if no clear "today")
- [<item title>](<link>) — <one-line summary>
- ...

**This week**
- [<item title>](<link>) — ...
- ...

**Older** (last 30 days, optional)
- ...

Cap each section at ~6 items. End with a one-line takeaway about the feed's overall direction this week ("Heavy on agents and tool-use this week"; "Quiet stretch — only 2 posts").

Multi-feed keyword filter

**Keyword digest** — feeds: <N>; keywords: <list>; matches: <count>

### <Feed name 1>
- [<item title>](<link>) — <published date> — <one-line context>
- ...

### <Feed name 2>
- [<item title>](<link>) — ...
- ...

Order feeds by match count (most-matching first). Within each feed, order items newest first.

Tone & failure modes

• Concise summaries — one sentence per item, distilled from the RSS summary field if it's available; otherwise just the title.
• Cite every item with a real <link> from the feed. Don't fabricate URLs.
• If a feed fails to parse, surface the error for that feed and keep going for the others. Don't abort the whole digest because one feed is malformed.
• If feedparser isn't installed, the script returns an error pointing at the skill's requirements: — surface that to the user and stop.
• If a feed has no recent items (last 30 days), say "Quiet — no items in the last 30 days" rather than padding with old content.
• For keyword search with zero matches, suggest broader synonyms or removing the filter entirely.
• No filler. The user is here for headlines, not commentary.
• If your host has no way to execute the script, say so plainly. Without RSS access, this skill cannot answer feed questions.