# Sitemap Desk > Audit an existing website's structure from its sitemap.xml, a crawl export or a URL list, then get > an architecture plan: the page hierarchy, the 301 redirect map, the navigation and the > internal-linking plan. https://sitemap-desk.skillsafe.ai/ Sitemap Desk is a web app on SkillSafe. The structure checks run free in the browser; the plan is a metered AI run (model alias gpt-terra) that is checked against the inventory afterwards. ## Input - A sitemap.xml (urlset; a sitemap index is recognised and the child sitemaps are asked for), a crawl export as CSV (an Address / URL / Loc column; Status Code and Inlinks columns are used when present), or one URL per line. - Optional: site name, site type (saas, content, ecommerce, docs, hybrid, local), goals, audiences, key pages, the current navigation, and a question. ## Free checks (in the browser, no AI, no upload) - Hosts and protocols: more than one site, www and bare host both present, http and https mixed. - Duplicates: the same URL twice, the same page under different case / trailing slash / index file, and the same slug under two folders. - Conventions: uppercase, underscores, file extensions, content selected by query strings, tracking parameters, dates in paths, numeric or hash ids, slugs over 60 characters or 8 words, spaces, non-ASCII, stray hyphens, and a mixed trailing-slash policy. - Hierarchy: pages deeper than the typical maximum for the site type (saas 3, content 3, local 2, ecommerce 4, docs 4, hybrid 4 - a house guideline), folders that hold pages but have no page of their own, top-level sections that mean the same thing (/product and /features), more than 12 top-level sections. - Linking: orphaned pages (0 inlinks) when the export carries an Inlinks column; key pages missing, more than 3 levels down, or orphaned. - Coverage: utility and search-result URLs in the inventory; pages a site of this type usually has. - Sitemap protocol: at most 50,000 URLs and 50 MB uncompressed per sitemap file (sitemaps.org protocol; Google Search Central), absolute URLs in , and the note that Google ignores and (Google Search Central, "Build and submit a sitemap"; checked 2026-09-26). Free exports: the prescan as Markdown and a per-URL audit CSV (url, path, depth, problems, inlinks, status). ## The plan (metered) Status: sound, tidy_up or restructure. Findings tied to the prescan flags, the proposed hierarchy (ASCII tree, URL map table, Mermaid sitemap; "/*" collection nodes stand for many detail pages), the 301 redirect map (exported for nginx, Apache RedirectMatch, Netlify _redirects, Vercel vercel.json - which uses 308 - and CSV), the header navigation with a separate CTA, footer columns, breadcrumbs, hub and spoke pages, cross-section links with anchor text, a placement for every orphan, and next steps. The reply is reconciled: every flag answered, status no looser than the flags left standing, proposed URLs obey the plan's own URL rules and mirror their parent, pages not marked new exist in the inventory, every redirect starts from a URL in the inventory and lands in the plan with no chains or loops, every old URL sent is kept or redirected, key pages within three levels, every orphan placed. A migration sheet CSV then lists every URL read - not only the ones sent - with where it goes under the plan (kept, redirected with wildcards expanded, or UNPLACED), ready to re-crawl after launch. Run the prescan again on the post-launch sitemap and it reports, against the saved plan, which redirected URLs are still listed as pages and which planned pages are not there yet. ## API Programmatic use: https://sitemap-desk.skillsafe.ai/api.html (task "architect"; fields site_name, site_type, goals, audiences, key_pages, current_nav, inventory, facts, question). ## Source Derived from the agent skill @coreyhaines31/site-architecture (https://skillsafe.ai/skill/@coreyhaines31/site-architecture; github.com/coreyhaines31/marketingskills, MIT - licence text at https://sitemap-desk.skillsafe.ai/LICENSE-COREYHAINES31-MARKETINGSKILLS.txt).