hanzo.ai/e2e/gates/agents.spec.ts

import { test, expect } from '@playwright/test'
import { pages, read } from './export'
import { readFileSync } from 'node:fs'
import { PROMPT } from '../../lib/agents'

/**
 * WHAT AN AGENT READS ABOUT HANZO, READ BACK OFF THE EXPORT.
 *
 * The prompt is one copy (lib/agents.ts) drawn on /agents and written into
 * llms.txt; the crawlers are named in robots.txt; the JSON-LD parses and says
 * what the company is; the API catalog is a linkset. Each of these is a file a
 * machine reads and a person never opens, so a regression in any of them is
 * invisible on the page.
 */

/** Every JSON-LD block in one exported page, parsed. */
function ld(html: string): unknown[] {
  return [...html.matchAll(/<script type="application\/ld\+json"[^>]*>([\s\S]*?)<\/script>/g)].map((m) => JSON.parse(m[1]))
}

/** Every node of every block, with `@graph` flattened. */
function nodes(blocks: unknown[]): Record<string, unknown>[] {
  return blocks.flatMap((b) => {
    const o = b as Record<string, unknown>
    return Array.isArray(o['@graph']) ? (o['@graph'] as Record<string, unknown>[]) : [o]
  })
}

test('llms.txt opens with the positioning, then For agents, the prompt and Company', () => {
  const txt = read('llms.txt')
  expect(txt.startsWith("# Hanzo\n\n> The AI operating system for business. Building business software since 2014. Techstars '17.")).toBe(true)
  const heads = [...txt.matchAll(/^## (.+)$/gm)].map((m) => m[1])
  expect(heads.slice(0, 4)).toEqual(['For agents', 'The prompt', 'Company', 'Pages'])
  expect(txt, 'the prompt is lib/agents.ts, byte for byte').toContain(PROMPT)
  expect(txt).not.toContain('[email protected]')
})

test('robots.txt names the AI crawlers and holds them to the same Disallow lines', () => {
  const txt = read('robots.txt')
  for (const bot of ['GPTBot', 'ClaudeBot', 'anthropic-ai', 'PerplexityBot', 'Google-Extended', 'CCBot', 'Applebot-Extended']) {
    expect(txt, bot).toMatch(new RegExp(`^User-Agent: ${bot}$`, 'm'))
  }
  const groups = txt.split(/\n\n+/).filter((g) => /^User-Agent:/m.test(g))
  const disallow = (g: string) => [...g.matchAll(/^Disallow: (\S+)$/gm)].map((m) => m[1]).sort()
  expect(groups.length).toBeGreaterThanOrEqual(2)
  for (const g of groups) expect(disallow(g)).toEqual(disallow(groups[0]))
  expect(txt).not.toMatch(/Bytespider/)
})

test('the root JSON-LD parses and says what Hanzo is, with no search box it cannot answer', () => {
  const graph = nodes(ld(read('index.html')))
  const org = graph.find((n) => n['@type'] === 'Organization')
  expect(org).toBeTruthy()
  expect(org!.foundingDate).toBe('2014')
  expect(org!.description).toBe("The AI operating system for business. Building business software since 2014. Techstars '17.")
  expect((org!.founder as { name: string }[]).map((f) => f.name)).toEqual(['Antje Worring', 'Zach Kelling'])
  const site = graph.find((n) => n['@type'] === 'WebSite')
  expect(site).toBeTruthy()
  expect(site!.potentialAction, 'no SearchAction: /search reads no query').toBeUndefined()
  const api = graph.find((n) => n['@type'] === 'WebAPI')
  expect(api?.url).toBe('https://api.hanzo.ai/v1')
  expect(api?.documentation, 'the agent manifest').toBe('https://hanzoskills.com/hanzo.md')
  expect(api?.provider).toEqual({ '@id': org!['@id'] })
})

test('/agents carries the prompt and states its questions as FAQPage', () => {
  const html = read('agents.html')
  const faq = nodes(ld(html)).find((n) => n['@type'] === 'FAQPage')
  expect(faq).toBeTruthy()
  const asked = (faq!.mainEntity as { name: string }[]).map((q) => q.name)
  expect(asked).toContain('What is free?')
  expect(asked).toContain('How does an agent pay?')
  for (const q of asked) expect(html, `the question "${q}" is drawn on the page`).toContain(q)
  expect(html).toContain('href="https://docs.hanzo.ai/docs/agents"')
  expect(html).toContain('href="/agents/sdk"')
})

test('/.well-known/api-catalog is an RFC 9727 linkset that names the OpenAPI document', () => {
  const doc = JSON.parse(read('.well-known/api-catalog')) as { linkset: { anchor: string; 'service-desc'?: { href: string }[] }[] }
  const api = doc.linkset.find((l) => l.anchor === 'https://api.hanzo.ai/v1')
  expect(api?.['service-desc']?.[0]?.href).toBe('https://api.hanzo.ai/v1/openapi.json')
})

test('every page names llms.txt (alternate and describedby), skill.md, the agent start page, the API catalog and the OpenAPI document in its head', () => {
  const heads = [
    '<link rel="alternate" type="text/plain" href="/llms.txt" title="llms.txt"/>',
    '<link rel="describedby" type="text/plain" href="/llms.txt"/>',
    '<link rel="alternate" type="text/markdown" href="/skill.md" title="skill.md"/>',
    '<link rel="help" href="/agents" title="For agents: get an API key in one call"/>',
    '<link rel="api-catalog" href="/.well-known/api-catalog"/>',
    '<link rel="service-desc" type="application/json" href="https://api.hanzo.ai/v1/openapi.json"/>',
  ]
  // /email/* are the bodies of the emails the site sends: a mail client reads them, not an agent.
  const shipped = pages().filter(({ route }) => !route.startsWith('/email/'))
  expect(shipped.length).toBeGreaterThan(100)
  const missing = shipped.flatMap(({ route, file }) => {
    const head = readFileSync(file, 'utf8').split('</head>')[0]
    return heads.filter((h) => !head.includes(h)).map((h) => `${route}: ${h}`)
  })
  expect(missing.slice(0, 20), `${missing.length} heads miss a pointer`).toEqual([])
})

test('/agents prints the headless script, and the Developers menu and footer lead with it', () => {
  const html = read('agents.html')
  expect(html).toContain('id="headless"')
  expect(html).toContain('/v1/iam/web3/verify')
  const home = read('index.html')
  expect(home.match(/href="\/agents"[^>]*>(?:<[^>]+>)*Agents: get an API key in one call/g)?.length ?? 0).toBeGreaterThanOrEqual(1)
})