Build #19 · Week 19 · Launched Oct 9, 2026
LiveCrawlKind
You built the website. Now fix what you missed. A $49 audit-and-fix kit for Claude Code: 23 checks for crawl blockers, missing metadata, and unclear content, with every fix shown as a diff you review in your own project. The first website it fixed is the one you are reading.

The idea
What it is & why I built it
On October 5, a reader audited this site and sent me the result: the raw HTML of sideroad.co had exactly one crawlable link to my builds per page, the homepage still said "Week 1 of 52, Building now" to any crawler that did not run JavaScript, and seventeen weeks of hub equity had passed almost nothing to the seventeen product sites. I fixed it that evening with a small pre-render script. Then I kept thinking about how many other AI-built sites are sitting in exactly that state, looking fine to their owners and half-invisible to crawlers.
CrawlKind is that evening turned into a product. It is a website audit-and-fix kit for Claude Code: a scan runs 23 checks for crawl blockers (noindex tags, robots.txt problems, redirect chains, broken links), discovery issues (sitemaps, canonicals, orphan pages), metadata (titles, descriptions, Open Graph, structured data), and whether a first-time visitor can tell what the page is for. Claude Code then turns the findings into fixes you review as diffs in your own project before anything lands.
The build
How it was built
The kit is a Claude Code skill plus one Python script, standard library only, 52 files total. It has three modes: audit, fix, verify, and the verify step names its method on every finding so you can check the work. It ships with a sample audit of a fictional 13-page site, labeled as fictional, with the real report the script produced for it.
The safety rules took as long as the checks. The script scans one host only, refuses private addresses, caps at 50 pages, and identifies itself as CrawlKind. The skill never deploys anything, never invents business facts to fill a meta description, and shows every proposed change as a diff first. Before launch, a second agent reviewed the code and found 12 bugs, robots.txt group parsing, relative links, sitemap encoding, all listed in the changelog. The last test was running it on its own site: the only finding was the deliberately hidden delivery page. It sells through build.sideroad.co, the same checkout BuiltByBadge uses.
The money path
Where the revenue opportunity is
$49 one-time, the highest price of the portfolio's four Claude-skill products; TextScrubr, JotThread, and BuiltByBadge all sit at $29. The buyer is anyone who shipped a site with AI and has not checked what crawlers actually see, a group that grows every week, and the proof the problem is real is not hypothetical: this site had it, the fix is documented in the journal, and the before-and-after link counts are public. The funnel is the sample audit on crawlkind.com, a launch post to the newsletter (drafted, publishing next week), and search content over time. Launch-day honesty: zero kits sold, and the first real test is whether "audit kit for Claude Code" reads as a category people already know they need.
The numbers
Analytics so far
The recap
Lessons & results
Four days from finding a problem on my own site to selling the fix for it, the shortest idea-to-product gap of the nineteen builds. The open question is whether buyers recognize the category before they get audited themselves. First kit sale, launch-post results, and the verdict land here.