My "push succeeded" content pipeline had been silently failing to publish for 13 days

I write technical posts through an agent-assisted pipeline: draft, edit, then push to a git repo connected to a hosted blog platform through a GitHub integration. Every step logged success. git add, commit, push - all green, every time.

What I hadn't done in a while was check the actual public page. When I finally did, the newest post visible there was 13 days old. Four newer pieces had been pushed since and never showed up.

The platform's own dashboard for the GitHub connection loaded and just sat empty - no error, no "disconnected" banner, nothing pending, nothing to click. I queried the platform's own API endpoint for that connection directly and got back an empty object. The link had died quietly somewhere in the middle, and every layer downstream of "git push succeeded" kept reporting success, because locally, it was true.

The bug in my own script was worse than the outage. After a successful push, it marked the post "published" and stamped the date - based entirely on the push succeeding, with nothing ever checking whether the platform had actually taken it. So the failure compounded: content sitting "published: true" in my own state file, invisible on the real site, for 13 days, with nothing flagging it the whole time.

The fix wasn't "be more careful going forward." It was moving the check outside the pipeline's own report entirely: before generating anything new, pull the platform's public RSS feed and confirm the last thing I pushed is actually in it. If it isn't, after a grace window, stop generating new content and write a loud alert instead of a quiet "published: true."

Same lesson I keep relearning at a different layer each time: a system that only checks its own intermediate steps will happily report full success on a process that shipped nothing.