How this site runs itself (n8n, Claude and one operator)
the actual stack behind arbhaveshpanse.com: markdown as the CMS, git as the undo button, and a workflow that publishes scheduled posts with no one at the keyboard.
this site is the proof, not the pitch. everything in this post is running on the domain you are reading it on, right now, and you can check the timestamps if you don’t believe me.
The stack, plainly
markdown files in a git repo are the entire CMS. every page, post and playbook is a file. Claude writes and edits the files directly, git tracks every change with a real diff, and a push to the main branch is the only publish button that exists. there is no dashboard, no login screen, no separate staging environment to keep in sync. the repo is the source of truth.
n8n runs the parts that need to happen on a clock instead of on a click. twice a day, a scheduled workflow checks every blog post’s publish date against today, finds anything that should be live but hasn’t rebuilt yet, and pushes an empty commit to trigger the deploy. that is the whole trick. a post can carry a date three months out, sit committed and gated behind that date, and go live on its own without me doing anything on the day it publishes.
Why this exists instead of a scheduling plugin
most CMS scheduling is a feature bolted onto a database. this is a two line check against files that already exist: has the date arrived, and has nothing published since. that second condition matters more than it sounds. an earlier version of this workflow fired on “is a post dated today,” which meant a missed run just silently skipped that post forever. asking “is a post’s date owed” instead means a dropped run gets picked up by the next one, twice a day, with no post ever lost to a flaky cron tick.
What Claude actually does versus what the workflow does
the workflow is dumb on purpose. it checks dates, it pushes commits, it does not touch content. all the judgment happens earlier: what to write, whether a claim is accurate, whether a number matches the canonical stats everywhere else on the site. a separate build step fails the entire deploy if a banned claim or a stale metric slips into anything published, which exists because a generator integration on this same site once reintroduced a fabricated case study three times before that check existed. the lesson was not “write better prompts.” it was “make the bad outcome physically impossible at build time,” which is a much better trade.
What this replaces, in hours
before this ran, a content calendar meant a person checking a spreadsheet and manually flipping a post live on the right day, which is exactly the kind of task that gets forgotten during a busy week. now a post gets written once, dated once, and the only thing left to do is nothing. that gap, between writing content and shipping it reliably on schedule, is most of what growth automation actually is. it is rarely glamorous. it is almost always a scheduling problem wearing a more exciting name.
if you want the same pattern applied to reddit specifically, and where it stops, that’s the n8n + Claude growth automations playbook.