An n8n workflow that scrapes a site, clusters its pages with an AI prompt, and writes the gaps to Google Sheets. What works, and what still needs cleanup.
Claude Code saw duplicate content, applied the textbook fix, and broke every TIL post on this site. The bug wasn't the model. It was where I kept my decisions.
Building 404.astro isn't enough. Without not_found_handling in wrangler.json, unknown URLs on a Workers site return a bare response instead of your page.
A static llms.txt goes stale. An Astro endpoint that reads your content collections rebuilds it every deploy. One gotcha: public/ silently shadows it.
A Bash rewrite hook pointed at a CLI that wasn't installed. Every shell command in every Claude Code session failed until Claude routed around it.
Most calls to businesses never reach a live person, and callbacks come hours too late. What that costs you, and how an AI voice agent closes the gap.
