Problem.
I wanted a watch spec reference site I'd actually use, plus a forcing function to learn full-stack ops with Claude Code as a daily collaborator. Existing references either lean commerce-coded or read like scraped catalog dumps — neither serves someone who just wants to compare a Tudor BB58 against an Omega Seamaster Planet Ocean side-by-side.
The constraints: solo build, deployed on the same VPS as everything else, no SaaS dependencies, automated social presence so the site has reach without me posting daily.
Approach.
Vite + React 19 frontend, hand-built per-brand scrapers (each marque has its own quirk — Panerai needs Scrapling stealth headers, Tudor wants alpha-aware image regen, Omega's CDN serves AVIF by default unless you ask for PNG). Claid.ai image pipeline standardizes the watches into a 4% padding ring, then post-crops vertical fill so bracelets touch top and bottom edges. All driven by Node hardened behind PM2.
The IG and X posters are independent cron jobs sharing the same catalog. The IG poster solved a Meta API quirk (their client_credentials grant returned 101 errors on two separate apps) by sidestepping to a never-expiring Page access token. The X poster runs at 13:00 UTC; the IG poster at 17:00 UTC, with a 30-day cooldown and 24-hour cross-exclusion so the same watch never appears on both platforms in the same day.
Brand list, watches, and post history are all flat JSON. No database. The whole system fits in a single ecosystem.config.js.
Outcome.
51 watches across 31 brands. Daily posts running unattended for weeks from @SuperSpecsWatch on X and @superspecs.watch on Instagram. First IG post hit the catalog automatically — Omega Seamaster Planet Ocean 600M, no manual intervention.
Site live at superspecs.watch with 99.9% uptime since launch. The Meta token sidestep is documented in project memory for the next time anyone hits Meta's client_credentials wall — saved future-me hours.
Gallery.
