An automatic proofreader for a whole website. One command reads every page and lists what should never go live: a title cut in the middle, a paragraph copied from another page with the numbers changed, an empty section, a link that leads nowhere, a figure with no source. Sixteen checks, thirteen seconds on 1,660 pages, and an eight-page PDF that explains each one. Free and MIT, nothing leaves your computer. It reads Markdown files: Astro, Next and MDX sites, not WordPress. npx @operstack/gates
Hi everyone. I run a pipeline that publishes to fifteen content sites, some of them past a thousand pages. Every one of these sixteen checks exists because that exact defect shipped at least once and the build log said nothing.
The one that hurt most: 121 pages sharing a closing paragraph with the numbers swapped. My duplicate checker compared exact strings, so it never saw it. Google's spam policy has a name for that pattern.
So the gates compare paragraphs with the digits masked. They also compile every MDX file up front (eleven-minute builds that fail on file 1,214 are not a workflow), check that llms.txt still matches the site, and flag any yield, occupancy or nightly-rate figure that has no source in the same paragraph.
It is free and MIT. npx @operstack/gates in any Astro or MDX project. The package ships with a fictional demo site so you can see every gate fire before you point it at your own corpus: npx @operstack/gates --demo
Happy to answer anything about the checks or the thresholds. They are all in one JSON file, and I would rather hear which ones are wrong than defend them.
One more thing for anyone whose site is not built from Markdown files: this package will not help you, but the free check on our site works on any site and installs nothing. It tells you whether assistants can read, understand and quote you: https://oper-stack.com/ai-visibi...
Hi everyone. I run a pipeline that publishes to fifteen content sites, some of them past a thousand pages. Every one of these sixteen checks exists because that exact defect shipped at least once and the build log said nothing.
The one that hurt most: 121 pages sharing a closing paragraph with the numbers swapped. My duplicate checker compared exact strings, so it never saw it. Google's spam policy has a name for that pattern.
So the gates compare paragraphs with the digits masked. They also compile every MDX file up front (eleven-minute builds that fail on file 1,214 are not a workflow), check that llms.txt still matches the site, and flag any yield, occupancy or nightly-rate figure that has no source in the same paragraph.
It is free and MIT. npx @operstack/gates in any Astro or MDX project. The package ships with a fictional demo site so you can see every gate fire before you point it at your own corpus: npx @operstack/gates --demo
Happy to answer anything about the checks or the thresholds. They are all in one JSON file, and I would rather hear which ones are wrong than defend them.
One more thing for anyone whose site is not built from Markdown files: this package will not help you, but the free check on our site works on any site and installs nothing. It tells you whether assistants can read, understand and quote you: https://oper-stack.com/ai-visibi...