hnrss

An RSS feed of Hacker News stories that:

• crossed a score threshold (default 125),
• have aged at least N hours (default 8, so top comments have settled), and
• include the top 4 comments inline (top-level #1, its top reply, top-level #2, its top reply).

Feed URL: https://elialbert.com/hnrss/feed.xml (also https://elialbert.github.io/hnrss/feed.xml).

RSS <link> points to the article; the HN comments link is the first thing inside the body.

How it works

GitHub Actions (cron, hourly at :07)
  → python fetch.py
      → Algolia HN search:   stories with points >= THRESHOLD, aged in [MIN_AGE, LOOKBACK]
      → HN Firebase API:     for each new story, walk kids[] in HN display order
                             to pick top comment + top reply + second comment + its top reply
      → write feed.xml + state.json
  → commit & push if changed
GitHub Pages
  → serves feed.xml from the repo root

Two APIs because they each do one thing better:

• Algolia (hn.algolia.com/api/v1/search) — easy numeric filters (points, created_at_i). Great for discovery.
• HN Firebase (hacker-news.firebaseio.com/v0/item/{id}.json) — kids is in HN display order (HN's ranking). Algolia's children array is creation-time order, which surfaces wrong "top" comments.

Files

File	Purpose
fetch.py	The whole generator. Stdlib only — no requirements.txt.
feed.xml	The published RSS feed. Regenerated each run.
state.json	Story IDs we've already added, with their cached rendered HTML. Acts as the dedup set and lets us regenerate the feed without re-fetching.
.github/workflows/update.yml	Hourly cron + commit + push.

Tuning

Constants at the top of fetch.py:

SCORE_THRESHOLD = 125    # min HN points
MIN_AGE_HOURS   = 8      # stories must be at least this old (lets top comments settle)
LOOKBACK_HOURS  = 72     # how far back to scan for candidates
MAX_ITEMS       = 100    # how many to retain in feed

To change what comments get pulled, see pick_comments() and build_html().

Running locally

python3 fetch.py    # writes feed.xml + state.json in cwd

Stdlib only. First run on an empty state.json will hit the HN Firebase API once per candidate-story top-level kid (a few hundred requests for a backfill, ~6 per new story on incremental runs). No auth, no rate-limit issues in practice.

To force a clean rebuild: delete state.json and re-run. (Items older than LOOKBACK_HOURS won't be re-discovered though.)

Manual cron trigger

GitHub repo → Actions → "Update feed" → Run workflow. Useful if you've just changed SCORE_THRESHOLD and want the feed to reflect it without waiting for the next hour.

Pages config

Repo Settings → Pages: source = "Deploy from a branch", branch = main, folder = / (root). The whole repo is the site; feed.xml lives at the root.