Rss Digest

Agentic RSS digest using the feed CLI. Fetch, triage, and summarize RSS feeds to surface high-signal posts. Use when: (1) reading RSS feeds or catching up on...

Security Scan

The skill's requirements and runtime instructions are consistent with an RSS-digest tool that uses the 'feed' CLI; nothing requested looks unrelated to that purpose.

Purpose & Capability

Name/description align with requiring the 'feed' CLI and the install options (Homebrew formula and Go package) both produce the 'feed' binary the SKILL.md calls; the OPML import URL and feed commands match an RSS digest workflow.

Instruction Scope

Instructions are limited to feed CLI commands and fetching article URLs for fuller text. This is appropriate for an RSS digest, but it explicitly directs the agent to fetch external web content (and to parallelize requests when possible), which can increase network activity and exposure of fetched content to whatever web-fetching tool the agent uses.

Install Mechanism

Install options are Homebrew (odysseus0/tap/feed) or a Go package from github.com/odysseus0/feed. Both are reasonable for distributing a CLI; the Go install will build arbitrary code pulled from GitHub (moderate risk compared with a vetted binary package). If you care about provenance, review the Homebrew tap and the GitHub repo before installing.

Credentials

The skill declares no required environment variables, credentials, or config paths; runtime instructions do not request secrets or unrelated credentials.

Persistence & Privilege

always:false and no install-time behavior appears to require persistent elevated privileges or modification of other skills or system-wide agent settings.

Assessment

This skill is coherent for producing RSS digests, but note two practical risks: (1) installing the 'feed' CLI via the Go package builds code fetched from GitHub—review the repo or prefer the Homebrew formula if you trust that tap; (2) the instructions explicitly fetch full article URLs (and recommend parallelizing), so the agent will perform network requests and download external page content—be comfortable with that network activity and the sites being contacted. Because the skill is instruction-only, there is no shipped code to audit here; if you need stronger assurance, inspect the odysseus0/feed GitHub repo and the Homebrew formula before installing, and consider limiting the agent's autonomous network access or invoking the skill only when you initiate it.

Surface what's worth reading from RSS feeds. Requires feed CLI (brew install odysseus0/tap/feed).

Workflow

1. Scan — feed get entries --limit 50 for recent unread (title, feed, date, URL, summary). Auto-fetches if stale. If 0 results, run feed get stats — if 0 feeds, import starter set: feed import https://github.com/odysseus0/feed/raw/main/hn-popular-blogs-2025.opml and retry.
2. Triage — Pick 5-10 high-signal posts based on the user's prompt. If no specific interest given, prioritize surprising, contrarian, or unusually insightful pieces.
3. Read + Synthesize — For each picked entry, read the full content and summarize in 2-3 sentences. Prefer fetching the URL directly (e.g. WebFetch) if available — keeps full text out of context. Otherwise use feed get entry <id> to read the stored content. Parallelize when possible.
4. Present — Compile the summaries into a digest. Group by theme if natural clusters emerge.

Commands

feed get entries --limit N              # list unread entries (table)
feed get entries --feed <id> --limit N  # filter by feed
feed get entry <id>                     # read full post (markdown)
feed fetch                              # pull latest from all feeds
feed search "<query>"                   # full-text search
feed update entries --read <id> ...     # batch mark read
feed get feeds                          # list feeds with unread counts
feed get stats                          # database stats

Notes

• The entries table includes full URLs. Prefer fetching URLs directly (keeps full text out of your context window). Fall back to feed get entry <id> if you don't have a web fetch tool.
• Do NOT mark entries as read. The user decides what to mark read.
• Default output is table — most token-efficient for scanning. Avoid -o json.
• Filter by feed if too many entries: --feed <feed_id>.