← ClaudeAtlas

agent-readable-sitelisted

Make a marketing or product site legible to AI agents and prove it, with an llms.txt generated from the same data the pages render from, a markdown mirror under /md/ with Accept text/markdown negotiation that never catches browsers, a robots.txt Content-Signal line, Link headers, 404s that answer in the format the client asked for, and optionally an MCP server card with a read-only server behind it. Use this whenever the work touches llms.txt or llms-full.txt, Accept text/markdown or Markdown for Agents, Content-Signal or ai-train in robots.txt, agent readiness or "agent-ready", isitagentready.com or is-agentic.com scores, api-catalog or .well-known files, an MCP server on a website, JSON errors for unknown API paths, or making docs and pricing readable by ChatGPT, Claude or Perplexity. Reach for it even for "just add an llms.txt", because a hand-written index drifts within a month and nobody notices. Not for SEO in general, structured data alone, or an MCP server over an authenticated product API.
tanfust/skills · ★ 0 · AI & Automation · score 73
Install: claude install-skill tanfust/skills
# Agent-readable site An agent that lands on a marketing site gets a React shell, a 404 page it cannot parse, or a hand-written `llms.txt` that lists a pricing page you rebuilt in March. None of that logs an error. The site looks fine in a browser and is invisible or wrong to the thing that increasingly sends the traffic. This skill is the set of surfaces a public site publishes for agents, the constraints that make them correct rather than merely present, and a script that asserts the whole contract against a running server. Specifications and their sources are in `references/SURFACES.md`; two of the surfaces are drafts and the file says which. Framework-specific constraints met in production are in `references/NEXTJS.md`. The executable check is `scripts/verify-agent-surface.sh`. ## Order of work 1. Decide what agents may and may not do, and write it down 2. Generate `llms.txt` from the data the pages already use 3. Publish a markdown mirror and negotiate on `Accept` without catching browsers 4. State your position in `robots.txt` with Content-Signal 5. Advertise the surfaces in `Link:` headers and answer 404s in the client's format 6. Optionally, an MCP server card with a server that actually answers 7. Verify with the script and the two scanners ## 1. Decide what agents may do Before publishing anything, write `/auth.md` (and mirror it at `/.well-known/auth.md`): what is public, what needs a human (checkout with a card, newsletter sign-up behind bot detection), and