Publish The AI Dispatch

Generate and publish a daily edition of The AI Dispatch — an AI news broadsheet styled with the newspaper design system from styles/newspaper.html.

Arguments: $ARGUMENTS

publish (default, or no args): Full pipeline — research, write, deploy
preview: Research only — show what stories would be covered, don't write HTML
regenerate YYYY-MM-DD: Redo a specific day's edition

Phase 0 — Setup

Parse arguments to determine mode (preview, publish, or regenerate).
Set today's date: YYYY-MM-DD format (or the specified date for regenerate).
Read styles/newspaper.html to extract the full CSS design system.
Read news/CLAUDE.md for section conventions and template structure.
Check if news/{date}/index.html already exists (skip if publish, overwrite if regenerate).
Read up to 3 most recent editions from news/*/index.html for deduplication:
- Extract headlines (<h2>, <h3> text)
- Extract source URLs (href="http...")
- Build a dedup set of covered topics and links

Phase 1 — Research (5 parallel sonnet subagents)

Launch 5 sonnet subagents in parallel, each using WebSearch:

Agent 1: AI Industry News

Search for the latest from Claude/Anthropic, OpenAI, Google AI/DeepMind, Meta AI, xAI, Mistral, and general AI industry news. Focus on product launches, policy changes, funding rounds, partnerships.

Agent 2: Open Source & Developer Tools

Search for new AI developer tools, open-source model releases, MCP servers, framework updates, and notable library releases.

Agent 3: AI Research & Papers

Search for AI research breakthroughs, new benchmarks, notable papers (arXiv), safety research, and technical achievements.

Agent 4: Community, Culture & Media

Search for AI community discussions, notable AI YouTube videos, podcast highlights, opinion pieces, regulatory developments, and cultural impact stories.

Agent 5: GitHub Trending

Search for today's top trending GitHub repositories — not just AI/ML, but across all languages and topics. Use WebSearch to find GitHub trending pages, aggregator sites (e.g., trendshift.io, ossinsight.io), and dev community posts about trending repos. Return the top 5-8 repos with: repo name, author, language, stars/growth, and a 1-sentence description. Include the GitHub URL for each.

Agents 1-4 return: 5-8 ranked story summaries with:

Headline
2-3 sentence summary
Source URL(s)
Significance rating (1-5)

Agent 5 returns: 5-8 trending repos with:

Repo name and author
Language and star count/growth
1-sentence description
GitHub URL

Phase 1.5 — Deduplicate

Compare all research results against the dedup set from Phase 0:

Drop any story whose headline or URL matches recent coverage
Flag stories that are follow-ups to recent coverage (can still run, but note the prior coverage)

If mode is preview: Stop here. Display the curated story list to the user and exit.

Phase 2 — Editorial Planning

This is where Claude exercises editorial judgment. Instead of rigid required sections:

Rank all stories by significance, newsworthiness, and reader interest.
Pick a lead story — the single most important story of the day.
Choose layout components based on what the day's news warrants:
- .card-lead — Lead/feature stories (1-2 per edition)
- .card-column — Standard column stories (3-6 per edition)
- .card-brief — Quick-hit briefs (2-5 per edition)
- .card-feature — Boxed feature with label (0-2 per edition)
- .pull-quote — Notable quotes from the day (0-2)
- .data-table — Data/benchmark comparisons (0-1)
- .alert-breaking — Only if genuinely breaking news
- .cols-2, .cols-3, .cols-2-1 — Column layouts as appropriate
Plan page structure — masthead, then sections in order of importance.
GitHub Trending table — Always include a .data-table near the end of the edition showing 5-8 trending repos with columns: Repo, Language, Stars/Growth, Description. Each repo name links to its GitHub URL.
Only constants: masthead with date, at least one lead story, source attribution on every story, GitHub Trending table, footer with archive link.

Output a structured editorial plan before building.

Phase 3 — Build (2-3 sequential opus subagents)

Agent 1: HTML Skeleton + Lead

Create news/{date}/index.html
Full inlined newspaper CSS (adapted from styles/newspaper.html)
Google Fonts: Playfair Display, Libre Baskerville, PT Sans Narrow
Masthead with "THE AI DISPATCH" nameplate, date, volume/issue number
Volume = 1, Issue = count of existing editions + 1
Lead story and first half of planned sections
All external links: target="_blank" rel="noopener"
Bylines with source attribution on every story

Agent 2: Remaining Sections + Footer

Read the file from Agent 1
Add remaining sections per editorial plan
Footer with:
- Archive link: <a href="../index.html">AI Dispatch Archive</a>
- Back to elevator: <a href="../../index.html">Back to Lobby</a>
- Generated-by credit line
Verify all links work, no broken references

Agent 3: Polish (optional, if needed)

Read complete file
Verify responsive design (check @media rules)
Ensure consistent newspaper styling throughout
Check that every story has source attribution
Verify no duplicate stories within the edition

Target size: 800-2000 lines per edition.

Phase 4 — Index & Ship

Read news/index.html (the archive hub page).
Add new entry at top of the archive list with:
- Date
- Lead headline
- Brief summary (1 sentence)
- Link to {date}/index.html
Commit with message: Publish AI Dispatch for {date}: {lead headline}
Push using: git config --global credential.helper store && echo "https://GGPrompts:$(gh auth token --user GGPrompts)@github.com" > ~/.git-credentials && git push origin main

Edition HTML Template Reference

Each daily edition follows this structure:

<!DOCTYPE html>
<html lang="en">
<head>
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <title>The AI Dispatch — {Month Day, Year}</title>
    <link rel="preconnect" href="https://fonts.googleapis.com">
    <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
    <link href="https://fonts.googleapis.com/css2?family=Playfair+Display:ital,wght@0,400;0,700;0,900;1,400;1,700&family=Libre+Baskerville:ital,wght@0,400;0,700;1,400&family=PT+Sans+Narrow:wght@400;700&display=swap" rel="stylesheet">
    <style>
        /* Full newspaper CSS inlined from styles/newspaper.html */
        /* Adapted: masthead says "THE AI DISPATCH" */
    </style>
</head>
<body>
    <div class="container">
        <header class="masthead">
            <div class="masthead-top">
                <span>Volume 1, No. {issue}</span>
                <span>{date}</span>
                <span>AI News Daily</span>
            </div>
            <h1 class="masthead-nameplate">The AI Dispatch</h1>
            <p class="masthead-motto">"All the AI News That's Fit to Compile"</p>
            <nav class="masthead-nav">
                <a href="../index.html">Archive</a>
                <a href="../../index.html">Lobby</a>
            </nav>
        </header>

        <!-- Editorially chosen sections here -->
        <!-- Use newspaper components: .card-lead, .card-column, .card-brief, etc. -->
        <!-- Every story gets a .card-byline with source attribution -->

        <footer class="footer">
            <p><a href="../index.html">AI Dispatch Archive</a> · <a href="../../index.html">Back to Lobby</a></p>
            <p class="footer-credit">Generated by Claude · <a href="https://github.com/GGPrompts/htmlstyleguides">Source</a></p>
        </footer>
    </div>
</body>
</html>

Key Rules

Self-contained HTML: Each edition is a single file with all CSS inlined.
No JavaScript required: Editions are pure static HTML/CSS.
Source attribution: Every story must link to its source. Never fabricate URLs.
Editorial freedom: Claude decides which sections/layout to use. No rigid template.
Deduplication: Never repeat a story that ran in the last 3 editions.
External links: Always target="_blank" rel="noopener".
Responsive: Include mobile breakpoints from the newspaper style guide.