# Agent Ready POC

> Proof-of-concept that an autonomous AI agent — Claude, running inside Claude Code — can ship a complete, modern, secure, accessible, agent-ready static site end-to-end on Cloudflare, with a human operator's role limited to approval at eight checkpoints reserved for human judgment.

> The brand is openly the build process. There is no fictional founder. Service pages describe concrete AI implementation patterns we would build for a real skilled-trades client, supported by real industry data (BLS, NFIB, US Chamber, ServiceTitan, Goldman Sachs).

## Pages

- [Home](/): What this site proves; three services we would build for a real client.
- [About](/about/): What "agent-ready" means; how this site was built; the methodology.
- [Services](/services/): Three deliverables — dispatch automation, estimate acceleration, field-tech copilot.
- [Dispatch automation](/services/dispatch-automation/): Confidence-Routed Intake Pattern; phone-and-web intake → AI routing → human exception review.
- [Estimate acceleration](/services/estimate-acceleration/): Multimodal Draft Loop; photo/voice/text capture → model drafts estimate → estimator approves.
- [Field-tech copilot](/services/field-tech-copilot/): Three-Modality Capture; voice + camera + position with offline-first sync.
- [Case studies](/case-studies/): The actual iteration log of how this site was built. iter-0-baseline through iter-3-content with gate score deltas.
- [How we engage](/how-we-engage/): The eight-checkpoint methodology applied to a real client engagement.
- [Blog](/blog/): Long-form notes (post-publication).
- [FAQ](/faq/): Twelve questions about the project, the build, and applying the methodology elsewhere. FAQPage JSON-LD schema.
- [Contact](/contact/): founder@agentreadypoc.com; Cloudflare Email Routing; Turnstile-protected form gated on a configured site key.
- [Probe](/probe/): agent-readiness inspector — fetches and parses any site's discovery surfaces (llms.txt, Markdown, MCP, agent-skills, OpenAPI, OpenID, pricing) and reports a 7-point coverage board. Read-only, deterministic, zero-token.

## Verification

- Gate 1 (Cloudflare Agent Ready scanner): green across reported categories.
- Gate 2 (Lighthouse + axe-core + W3C HTML validator): 100/100/100/92 on desktop and mobile across all pages; 0 axe violations; 0 W3C errors. The 8-point SEO gap is a deliberate trade-off — the Cloudflare Content-Signal directive in robots.txt is flagged "Unknown" by Google's parser, and the project keeps the directive to clear Gate 1's Bot Access Control category.
- Gate 3 (content audit): 5–6 verifiable citations per service detail page; original named framework per service; FAQPage schema; passage-coherent H2 sections per published research on AI-surface content optimization.

## Optional

- [Sitemap](/sitemap.xml)
- [robots.txt](/robots.txt) — non-blanket AI bot rules: training-only bots disallowed (GPTBot, Google-Extended, anthropic-ai, CCBot, Applebot-Extended, Bytespider, FacebookBot); inference bots welcomed (ChatGPT-User, ClaudeBot, PerplexityBot, OAI-SearchBot).
- [llms-full.txt](/llms-full.txt) — concatenated content of every page.
- [/api/agent-info](/api/agent-info) — JSON summary of the organization, services, and discovery URLs. Wildcard CORS; GET-only.
- [/api/probe](/api/probe) — POST { "target": "<url>" } → JSON coverage report of that site's agent-ready surfaces. Read-only inspector: fetches and parses public discovery files; never calls, authenticates, or transacts. Powers /probe/.
- [/api/probe-mcp](/api/probe-mcp) — POST { "target": "<url>", "consent": true } → live MCP session (initialize + tools/list) with card-vs-live drift; action:"call" invokes ONE tool only if it is annotated read-only, non-destructive, and argument-free. Consent required; mutating/auth tools shown, never called. Powers the live MCP panel on /probe/.
- [Source repository](https://github.com/agent-ready-ai/agent-ready-poc) — full source, MIT licensed.