hn-morning-brief

## Morning Briefing ### Step 1 — Pull user interests from memory ``` memory_search("interests topics preferences technology news") ``` Do this first, before fetching stories — the results determine how stories are ranked. Extract any topics, technologies, or themes found. If nothing relevant comes back, fall back to HN ranking order. ### Step 2 — Fetch top HN stories ```bash python3 skills/hn-morning-brief/scripts/fetch_hn.py --limit 20 ``` (Path is relative to the project root — openclaw installs this skill at `skills/hn-morning-brief/`.) Returns JSON with: `title`, `article_url`, `hn_url`, `domain`, `author`, `points`, `num_comments`. ### Step 3 — Rank and filter Score each story by combining two signals: - **Relevance to user interests** (from memory) — a story the user cares about is worth more regardless of score - **HN points** — use as a tiebreaker and quality signal when interests are unclear Surface the 8–12 highest-scoring stories. If memory search returned no clear interests, rank by `points` only. ### Step 4 — Present briefing ``` ## HN Morning Brief — {today's date} {N} stories picked for you 1. **{Title}** `{domain}` · ⬆ {points} · 💬 {num_comments} {one-line context or why this is interesting} → [Article]({article_url}) · [HN Discussion]({hn_url}) 2. ... --- Say "dive deeper into #N" or "tell me more about [title]" to get a full summary. ``` --- ## Diving Deeper When the user picks a story: 1. **Fetch and summarize the article** — read the article URL and write a 3–5 sentence summary of the key points. Do this even if the user just says "more on #3" — they want the content, not just the link. 2. **Show both links:** - Article: `{article_url}` - HN Discussion: `{hn_url}` (often where the most interesting debate happens) 3. **Offer to go further:** "Want me to search for more context on this?" ## Gotchas - `article_url` is the **original article**. `hn_url` is the HN discussion thread. Never swap them — linking to the HN page when the user wants the article is a bad experience. - If the article is a PDF or appears paywalled, say so and summarize from the title, domain, and any available description instead of silently failing. - If `memory_search` returns no clear interests, rank by `points` and don't guess — invented interests will surface irrelevant stories.