hey! i was testing an llms.txt validator against popular docs sites and vercel.com/llms.txt came back with 18 broken links out of 1733. the fun part: 8 of them are the same literal URL, https://vercel.com/docs/rest-api/undefined/no-summary, which looks like the generator interpolating undefined for rest-api entries that lack a slug or summary. the other 10 are pages that moved or were removed while llms.txt kept the old routes:
https://vercel.com/docs/organizationshttps://vercel.com/docs/ai-gateway/security-and-compliance/private-inferencehttps://vercel.com/docs/ai-gateway/models-and-providers/virtual-modelshttps://vercel.com/docs/vercel-firewall/vercel-waf/team-configurationhttps://vercel.com/docs/vercel-firewall/vercel-waf/trusted-ip-accesshttps://vercel.com/docs/vercel-firewall/page-integrityhttps://vercel.com/docs/vercel-firewall/security-plushttps://vercel.com/docs/networking/privatelinkhttps://vercel.com/docs/rest-api/domains-registrar/get-supported-tld-shttps://vercel.com/docs/rest-api/networking/configures-static-ip-s-for-a-projecthttps://vercel.com/docs/rest-api/undefined/no-summary (listed 8 times)quick check: curl -so /dev/null -w "%{http_code} " https://vercel.com/docs/rest-api/undefined/no-summary → 404
any AI agent that reads your llms.txt to answer vercel questions hits 404s on these, and the undefined ones suggest the generator needs a guard for missing fields. reproducible with npx llms-txt-check https://vercel.com (llms-txt-check, a checker i built for this; it validates deployed llms.txt files against what the site actually serves).
for context, i maintain docusaurus-plugin-copy-page-button (runs on 28 docs sites; react native and pnpm are the big ones, puppeteer too) so i spend a lot of time in this corner of docs tooling.