Overview
Surface what's worth reading from RSS feeds. Requires feed CLI (brew install odysseus0/tap/feed). Workflow 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. 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. 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. 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>.
SKILL.md file
Preview raw SKILL.md. Open the full source below. Scroll, inspect, then download the exact SKILL.md file if you want the original.
# rss-digest
RSS Digest
Surface what's worth reading from RSS feeds. Requires feed CLI (brew install odysseus0/tap/feed).
Workflow
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.
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.
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.
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>.
Preview raw SKILL.md. Open the full source below. Scroll, inspect, then download the exact SKILL.md file if you want the original.
# rss-digest
RSS Digest
Surface what's worth reading from RSS feeds. Requires feed CLI (brew install odysseus0/tap/feed).
Workflow
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.
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.
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.
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>.
Comments & Discussion
Add a comment