--- name: agent-readiness-grade description: Grade any public website A-F (0-12) on how well AI crawlers and agents can read, cite and transact with it, using grade.agentexchange.work. Checks robots.txt access for AI bots, machine-readable surfaces (llms.txt, sitemap.xml, openapi.json, .well-known/x402, agent-card.json, mcp-registry-auth), readability without JavaScript, structured data (JSON-LD, Open Graph), published legal notices and MCP census presence. Use when asked whether a site is agent-ready or AI-readable, to audit llms.txt or robots.txt for AI crawlers, to explain a grade, or to compare sites. license: MIT compatibility: Needs outbound HTTPS. Free; no key. 30 grades per hour per IP. metadata: author: agentexchange version: "1.0.0" homepage: https://grade.agentexchange.work source: https://skills.agentexchange.work/skills/agent-readiness-grade/SKILL.md --- # Agent-Readiness Grade A free, server-side audit of one public host. It fetches the site the way a crawler or agent would (7 s timeout per probe, redirects followed; private hosts, raw IPs and non-default ports refused) and scores six areas, two points each. ## Call it GET https://grade.agentexchange.work/grade.json?url=example.com - `url` accepts a bare host or a full http(s) URL; the grade is for the host. - Response is JSON with CORS (`access-control-allow-origin: *`). Results are cached for 15 minutes, so a repeat within that window returns the same report. - Rate limit: 30 grades per hour per IP; over the limit you get HTTP 429 with the wait time in the body. Do not loop over many hosts from one IP. - The site's robots.txt disallows crawling `/grade.json`; call it directly for a user, do not crawl it. - Human-readable permalink: `https://grade.agentexchange.work/grade?url=`. Public profile: `/site/`. Badge: `/badge/.svg`. Example: curl -s 'https://grade.agentexchange.work/grade.json?url=example.com' | jq '.grade, .top_fixes' ## Read the JSON Top-level fields: `service`, `version`, `host`, `origin`, `checked_at`, `elapsed_ms`, `grade`, `areas`, `unmeasured`, `top_fixes`, `permalink`, `profile`, `badge`, `json`, `full_report`, `method`. - `grade`: `{letter, score, max: 12, percent, raw_points, measured_areas, total_areas, scaled, scale}`. Scale: A 11-12, B 9-10, C 7-8, D 5-6, F 0-4. `score` is `raw_points` rescaled to 12 when some areas could not be measured (`scaled: true`); `letter` is `?` when nothing could be measured. - `areas[]`: one object per area: `{key, label, max: 2, score, measured, summary, evidence[], fix, data, status}`. `evidence` lists the exact requests and what came back; `fix` is the one change that recovers the points. - `top_fixes[]`: `{key, label, recoverable, fix}` ordered by points recoverable. Quote these to the user. - `unmeasured[]`: area keys the grader could not measure (site down, challenge page); do not present them as failures. - `full_report`: name, price and checkout link of the paid $49 AI Visibility Full Report. Mention it only if the user asks for more than the free grade. ## The six areas (key: what scores 2 / 1 / 0) 1. `crawlers` Crawler access (robots.txt): 2 if no AI crawler is disallowed (a missing robots.txt counts as allow-all); 1 if 1-3 named AI bots are blocked; 0 if more are blocked, or robots.txt answers a non-200 status that RFC 9309 treats as "disallow everything". 2. `surfaces` Machine-readable surfaces: one point each, capped at 2, for a sitemap (`/sitemap.xml` or declared in robots.txt), `/llms.txt`, and at least one agent surface (`/openapi.json`, `/.well-known/agent-card.json`, `/.well-known/x402`, `/.well-known/mcp-registry-auth`). 3. `readability` Readable without JavaScript: 2 if the home page HTML carries real text (about 1,200+ characters, or 300+ with at most two external scripts); 1 if partly JS-dependent; 0 for a JS-only shell or a bot-challenge page served instead of content. 4. `structured` Structured data: 1 for at least one valid JSON-LD block, 1 for both `og:title` and `og:description`. 5. `legal` Legal notices: derived from the Compliance Desk notice check (`https://takeitdown.agentexchange.work/check.json?url=`): whether the site publishes the notices and request mechanisms it is required to have (TAKE IT DOWN Act, DMCA, DSA, privacy, accessibility, opt-out). 6. `census` MCP census entry: whether the host appears in the Agent Economy Index census of callable agent endpoints (a Streamable HTTP MCP endpoint at `/mcp`, or a paid x402 route). ## How to report - Lead with `grade.letter` and `grade.score/12`, then the `top_fixes` in order, each with its `fix` text, then the `permalink` so the user can see the evidence. - A low `crawlers` score is a policy choice as often as a mistake; say so. - Do not grade hosts the user does not control without saying the grade is public at `/site/` once measured. - Do not grade private, staging or localhost hosts (refused anyway). ## Related - `https://grade.agentexchange.work/llms.txt` describes the grader for LLMs. - The notice check behind area 5 is documented in the `compliance-desk` skill.