hn-digest

HN Digest Skill

"What's on HN?" deserves better than thirty raw titles — the value is in the digest: what's leading, what the comment sections are actually arguing, and which of it the user cares about. Hacker News serves everything keylessly twice over: the official Firebase API (live items by id) and Algolia's HN search (query, date ranges, popularity). This skill knows when to use which, batches sanely, and summarizes discussions as positions, not vibes.

What This Skill Produces

• The digest — top stories with score, comments count, domain, and a one-line what-it-is
• Discussion reads — for a story: the top comment threads compressed into the actual arguments being made
• Topic searches — has-HN-covered-X, with dates and reception
• The commands — exact curls, rerunnable

Required Inputs

Ask for these if not provided:

• The mode — front page now, a specific story's discussion, or a topic search
• Appetite — top 5 headline-digest vs. deep read of one thread
• Their filter — "anything about AI/security/startups" turns a digest into a targeted one; worth asking when the user has an obvious beat

Framework: The Two APIs and the Digest Rules

1. Front page (official API): ids: curl -s "https://hacker-news.firebaseio.com/v0/topstories.json" → then per item curl -s "https://hacker-news.firebaseio.com/v0/item/{id}.json" → title, score, descendants (comment count), url, by, time. Fetch the top 10–15 ids only — not all 500. Also: beststories, newstories, askstories, showstories.
2. Search and story lookup (Algolia): curl -s "https://hn.algolia.com/api/v1/search?query=postgres&tags=story&numericFilters=points>50" — or by date: search_by_date. A story's full comment tree in one call: https://hn.algolia.com/api/v1/items/{id}. For topic questions Algolia is one request where Firebase is fifty.
3. Digest, don't dump: each story gets one line of what-it-actually-is (from the title and domain — fetch the linked article only if the user asks); lead with the 3–5 highest-signal items for the user's stated interests, then the rest compressed.
4. Comment sections are position maps: summarize a discussion as its distinct arguments ("top thread argues X; the main pushback says Y; a practitioner reports Z") with rough weight — never as "mixed reactions," which is always true and never informative. Attribute claims to "a commenter," not as facts.
5. Freshness and voice: scores are moving snapshots — timestamp the digest; and HN comments are opinions with usernames, quoted as such. The front page is also a specific community's taste — say "HN's take" not "the tech world's take."

Output Format

HN Digest — [time, user's zone]

[If filtered: the matching stories first.]

| # | Story | Score · Comments | What it is |

|---|---|---|---|

[Discussion mode: the position map — 3–5 argument threads, weighted, attributed]

[Search mode: matches with dates and reception]

Source: [HN Firebase API / Algolia HN search] · rerun: [exact curls]

Scores are live snapshots; comments are commenters' views, not facts.

Quality Checks

• [ ] Fetches were batched sanely (top 10–15, not 500 ids)
• [ ] Every story line says what the thing is, not just its title
• [ ] Discussion summaries name distinct positions with rough weights — no "mixed reactions"
• [ ] Comment claims are attributed, not laundered into facts
• [ ] The digest is timestamped

Anti-Patterns

• [ ] Do not dump 30 raw titles — the digest is the product
• [ ] Do not loop the Firebase API when Algolia answers in one call
• [ ] Do not present commenter claims as verified facts — attribute or drop
• [ ] Do not answer "what's on HN" from memory — the front page turns over in hours
• [ ] Do not editorialize the community's votes into objective importance — it's HN's taste, labeled as such