This repository builds the Story Computing Machine home page, library, story readers, and art gallery. The source repository owns the writing and artwork. This repository owns the public presentation and its pinned content/catalog.json publication snapshot (currently 222 stories).
The homepage is The Long Horizon, a scroll-driven WebGL exhibition built with three.js (vendored in static/vendor/). It opens on one landscape painting assembled from one tile for each of the 222 story covers. The larger portrait covers form a mosaic that crops to fill any screen size, and hover smoothly lifts and enlarges a cover; hovering names its story and clicking opens it. Cover tiles for stories created within the last 48 hours are spread through the upper rows and stand out through brighter color and warm edges. Scrolling releases the tiles and flies the camera through every story from the last 48 hours plus ten randomly chosen older stories. A single open storybook stays fixed behind the entire passage while the story art moves in front of it. The quiet parchment pages are cropped beyond the viewport without stretching, with darker areas behind the header and footer. Each stop is a location painting from that story's art files, broken into shards that lift from the open pages directly into a screen-filling landscape; the cover and prompt rise from the pages with it. Character reference sheets are reserved for the gallery and do not appear in the passage. The pieces return to the book as the camera passes. A shared book-cover-inspired location painting fills in for stories whose source package has no location art yet. Stops are ordered by the brightness of their skies, so the walk runs from morning into night and ends with the walk's covers hung beside a curator's note. Recent eligibility is calculated in the browser from createdAt, so it expires without a rebuild. Returning from a story restores the same walk and position. A visually hidden list of the stops keeps the walk usable from the keyboard and screen readers, and browsers without WebGL get a link to the library. The searchable grid lives at /library/; existing story URLs and /art/ are preserved.
The walk's art is prepared ahead of time because the published landscapes are 3 MB PNGs and WebGL needs same-origin textures. python scripts/sync_panorama.py [--repository PATH_TO_SOURCE] (requires Pillow) chooses a pool of 40 stories (the newest six with landscapes plus an even spread of the rest), downloads their art from the verified media index, and writes WebP derivatives, the mosaic atlas, and cover derivatives for the newest stories to static/panorama/ with a manifest in content/panorama.json. python -m scripts.sync_story_locations [--repository PATH_TO_SOURCE] prepares a same-story location and local cover for every story in content/story-locations.json, reusing the pool images and writing other derivatives to static/story-locations/ and static/story-covers/. Re-run both after importing stories or new art. Location derivatives prefer paintings with a -selected filename suffix and place preserved -original paintings last.
The guide API and client remain in the codebase but are not linked from the current homepage. The guide sets the reader's highest comfortable rating as starting context, asks up to four adaptive questions, and recommends three catalog stories through OpenRouter. Answers are kept in browser memory during the session and sent through the Worker to OpenRouter for each request; the Worker does not store them.
There is no category taxonomy or scheduled story classification. New stories become eligible when they are imported into content/catalog.json and the site is rebuilt. The guide's model ID is set by CHAT_MODEL in wrangler.jsonc (default openai/gpt-5-mini).
Ratings use the source collection's AO3 scale: General, Teen, Mature, and Explicit. The guide treats the reader's choice as the highest acceptable rating and filters stories on the server before asking the model to select any. content/source.json, content/covers.json, and content/ratings-source.json record the pinned revision, cover hashes, and rating counts. Onyx's source package does not yet have ratings.md; content/rating-overrides.json records its General rating, supported by the source editorial notes. To refresh from a reviewed source commit, run python scripts/sync_ratings.py --repository PATH_TO_SOURCE --revision COMMIT; pass --add-story SLUG for each new published story, oldest first. Refresh content/media-index.json from the public R2 index after its covers are published there.
Install Python and Node.js 24, then:
python -m pip install -r requirements.txt
npm ci
python -m unittest discover -s tests -p 'test_*.py'
npm run check
npm run test:ts
npm run build:siteRun npm run dev to view the site at http://127.0.0.1:8787/. An ignored .dev.vars file with OPENROUTER_API_KEY=... is needed only for guide API requests. npm run build:site:offline uses the checked-in art index when working without the live index. The build uses the R2 art index when available and falls back to content/media-index.json; _site/media-source.json records which it used.
The Worker serves the API and static _site assets together. It requires a Cloudflare account with the stories.rgbknights.com zone and a Worker secret named OPENROUTER_API_KEY. Never place that key in a public repository, a GitHub variable, or a client file. wrangler.jsonc configures a per-IP rate limit of 15 guide API calls per minute.
The .github/workflows/worker.yml workflow deploys a workers.dev preview when manually dispatched with the default preview target. Set the GitHub Actions secrets CLOUDFLARE_API_TOKEN, CLOUDFLARE_ACCOUNT_ID, and OPENROUTER_API_KEY first; the workflow uploads the last value as a Worker secret. The production deployment runs on pushes to main when the repository variable CLOUDFLARE_ENABLED=true, or through a manual production dispatch. The stories.rgbknights.com custom domain is configured in wrangler.jsonc and served by the Worker. GitHub Pages has been removed from this repository; .github/workflows/site-check.yml only builds and tests the site.
Cloudflare assets serve the existing story reader paths; the Worker handles / and /api/*. If OpenRouter is unavailable, the guide presents a link to /library/. The catalog and full story text are already public content; the built guide-catalog.json is served as a static asset.