The whole thing started in a single Claude conversation in April 2026. The brief was simple: build a daily newsletter website. The domain existed. GitHub existed. The content direction was forming. What was needed was a home that could handle daily publishing without recoding anything each time, had a real archive, and — critically — looked like it meant it.

The specific aesthetic ask was a nod to the Daily Prophet from Harry Potter: a broadsheet newspaper feel, but not old or parchment-y. Organic and computery. Matrix. Wires. The response to that brief became a complete Claude Code prompt used to scaffold the entire site — directory structure, config files, theme, sample posts, and deployment instructions in a single paste.

The prompt specified colors, every font by name, a CSS circuit-board SVG background texture at 5–8% opacity, and newspaper-style double-rule dividers with a faint glow. Delivery format was explicit: every file with its full path, no abbreviations, nothing left for the developer to figure out.

That first conversation also produced the page structure (homepage, archive, about, marketplace, subscribe), the _data/ads.yml approach for the Marketplace’s rotating fake cyborg-upgrade ads, and the first twelve ad copy samples — deadpan corporate voice selling neural lace, ocular upgrades, arm replacement therapy, and memory expansion modules. The Marketplace was never meant to be functional. It was always flavor text that committed to the bit.

The Daily Cyborg is not a both-sides publication. It has a specific position: conscious, deliberate human agency in the face of technological change is the goal. The name, the tone, the aesthetic, and the automation are all in service of that stance. Every design decision — from the CRT aesthetic to the three-weight editorial structure — is downstream of that premise.

The site runs on Jekyll, a static site generator, deployed via GitHub Pages at thedailycyborg.com. Every post is a markdown file in _posts/. Jekyll builds the HTML at deploy time — no server, no database, no CMS. The CNAME file ties the GitHub Pages deployment to the custom domain.

Posts use YAML front matter to carry metadata: layout, title, issue (a sequential integer shared across all editions), date, image, excerpt, and tags. The templates read these fields to render the masthead, issue label, themes line, and hero image — none of that content lives in the post body.

The visual language was not improvised in code — it was designed first, then handed off. A formal design handoff was produced (preserved in design_handoff_extracted/design_handoff_v1_muted_phosphor/) containing the full design system: palette, typography, CSS variable naming, spacing rules, and component layouts.

That handoff was imported into the Jekyll site. The design system has a name — v1 Muted Phosphor — and the reference files (shared.jsx, v1-muted-phosphor.jsx, v1-preview.html) served as the source of truth during implementation. The aesthetic is deliberately worn and low-fidelity: a técnicos-precise broadsheet printed on ancient hardware.

Each post displays a unique hero image representing the edition’s themes. Claude doesn’t generate images natively, so a separate tool was needed — and rather than hand-prompting each image individually (unsustainable at volume), a dedicated Gemini-powered web app was built: The Daily Cyborg’s Digital Administrative Assistant.

The app runs as a local React/TypeScript app (Vite) against the Gemini API, scaffolded in Google AI Studio. The input is a CSV file of keywords — topics drawn from the publication’s themes — uploaded to a queue and processed in bulk.

The workflow is a two-step pipeline:

There are currently 84 registered images in the pool, each tagged by topic and tracked in _data/images.json. A rotation algorithm checks the 10 most recently published posts and excludes those dates from the selection pool, preventing repeats in back-to-back editions.

The automation runs entirely through Claude’s scheduled task system — three separate tasks, each with a SKILL.md prompt file that Claude executes on a timer. No human intervention required on a normal publishing day.

The hopper is a pre-publish queue for content Jason wants to guarantee appears in a future edition. It’s backed by an Obsidian vault (_hopper/ with .obsidian/ config) so Jason can write and clip in Obsidian, then drop files into the appropriate folder. The scheduled task checks the hopper before searching the web and always prioritizes it.

After any hopper file is used, it moves to _hopper/archive/ with today’s date prepended. A hidden HTML comment at the bottom of every published post records what was used and which section it filled — invisible on the page, visible in the source.

The archive page (/archive/) doubles as a full-content search interface. Jekyll generates search.json at build time — a flat array containing every post’s title, URL, date, issue number, excerpt, full stripped content, and tags. The page fetches this on load and filters it client-side on every keystroke, matching against all four fields simultaneously. No server, no search API. Entirely static.

The override lives as a single block at the top of each SKILL.md prompt. Removing that block when Jason returns is the only change needed — nothing in the templates or downstream code is affected.

A standalone feature at /singularity-tracker/ pulls live probability data from Manifold Markets, Metaculus, and Kalshi on AGI arrival dates, alongside curated static entries from Kurzweil, Dario Amodei, and others. It uses manual fallback values for markets where live API access is blocked by CORS, cached for two hours in localStorage. Implemented in assets/js/singularity-tracker.js.