This site is fully automated. No one writes the digests by hand — every morning a GitHub Actions workflow runs the whole pipeline end to end, from fetching raw feed data to publishing the finished site. Here’s the full picture.

Architecture

flowchart TD
    subgraph schedule[" "]
        cron([⏰ GitHub Actions
daily cron · 5 AM EST])
    end

    subgraph aggregate["1 · Aggregate"]
        feeds[(config/feeds.yaml
curated feed list)]
        py[🐍 news-aggregator
fetch · filter 24h · dedupe]
        seen[(reports/seen.json
dedupe memory)]
        raw[/raw-YYYY-MM-DD.json/]
        feeds --> py
        seen <--> py
        py --> raw
    end

    subgraph synth["2 · Synthesize"]
        claude[🤖 Claude Code
read · rank · group · write]
        web([🌐 WebSearch
breaking news])
        post[/content/posts/
YYYY-MM-DD.md/]
        raw --> claude
        web --> claude
        claude --> post
    end

    subgraph publish["3 · Publish"]
        hugo[⚡ Hugo
build static site]
        pages[☁️ GitHub Pages]
        post --> hugo
        hugo --> pages
    end

    cron --> py
    pages --> reader([👀 You])

    style cron fill:#2563eb,color:#fff
    style claude fill:#d97757,color:#fff
    style reader fill:#16a34a,color:#fff

Step by step

1 · Aggregate

A small Python tool (news-aggregator) reads the feed list from config/feeds.yaml, fetches every feed, and keeps only entries from the last 24 hours. It tracks every URL it has already seen in seen.json, so the same story never shows up twice across days. The result is a single raw JSON file of the day’s new entries.

2 · Synthesize

Claude Code reads that raw JSON and does the editorial work: it skips marketing fluff and minor patch notes, groups articles that cover the same story, picks the day’s top story, sorts everything into themed categories, and assigns each item a relevance tier. It also runs a web search to catch breaking news the RSS feeds may have missed. The output is one Markdown post with front matter — the same format you’re reading now.

3 · Publish

Hugo builds the Markdown into this static site — archive, navigation, RSS feed, and all — and the workflow deploys it to GitHub Pages. Posts older than 60 days are pruned automatically to keep things current.

The whole thing is open source

Feed config, aggregator, Claude instructions, and site templates all live in one repo. Browse it, fork it, or suggest a feed on GitHub.