There's a new well-known path on the web. Cloudflare's agent skills discovery RFC puts a JSON index at /.well-known/agent-skills/index.json listing the SKILL.md files an agent can install from your domain, each with a sha256 digest so the agent knows when something changed. Cloudflare's Agent Readiness scanner already checks for it, and Mintlify generates one for every docs site it hosts.
We just published ours. The index lists five skills we use on real client work.
The five skills
content-negotiation-audit points your agent at any website, pulls the sitemap, samples ten pages, and requests each one twice: once as a browser, once with Accept: text/markdown. Then it grades the gap. Most sites fail immediately because they serve HTML no matter what you ask for. The interesting failures are further in: markdown routes that leak unrendered <Content> placeholders, or FAQ pages where every answer lives behind an accordion and never makes it into the markdown at all. Your agent reads a page that looks complete and is missing the exact content people ask about.
sitemap-audit turns a sitemap into an honest inventory: which templates repeat, which pages are one-offs, and which URLs are leaked drafts that should worry you. We run it before every rebuild proposal because the page type nobody mentioned is the one that blows the estimate.
cms-to-mdx-migration is the distilled version of how we migrated this very website off a headless CMS onto MDX files on disk: resumable migration scripts, rich text conversion gotchas, and the component mapping rules that prevent silent rendering failures. If you'd rather hand the whole migration over, that's a service we run.
sanity-schema-design and sanity-groq-conventions are the conventions we hold every Sanity project to: field templates with descriptions editors can actually use, and GROQ that stays readable past the tenth query.
All five are MIT licensed. Each is a single SKILL.md file, so there's nothing to unpack.
Point your agent at it
Browse the human-readable pages at robotostudio.com/skills or fetch the index directly:
Each entry carries a sha256 digest computed over the exact bytes served, so your agent can pin a version and notice when we ship a change.
Cloudflare's scanner does the same thing
The pattern is already how the graders grade. Cloudflare's scanner at isitagentready.com distributes its fix instructions as installable skills served from /.well-known/agent-skills/ on its own domain: every failing check links a SKILL.md that tells your agent how to fix it. The tool that scores agent readiness ships its entire playbook through the exact channel we just joined.
Why give it away
Because the moment someone runs content-negotiation-audit against their own site is the moment they find out their FAQ answers are invisible to AI. Some of those people will fix it themselves with the skill we just handed them. Some will want it done properly. Both outcomes are fine with us.
The site you're reading practices what the skills preach: every canonical URL on robotostudio.com answers Accept: text/markdown, the index digests match the served bytes, and /llms.txt documents the whole surface. We wrote up the general approach in AEO is just SEO with content negotiation.
Wow, I can't believe people are actually using these. Tell me if it worked: yo@robotostudio.com