nuxt-agent-ready-best-practices
starsforks
How to install nuxt-agent-ready-best-practices?
vinayakkulkarni/vue-nuxt-best-practices/nuxt-agent-ready-best-practices
$npx -y skills add vinayakkulkarni/vue-nuxt-best-practices --skill nuxt-agent-ready-best-practicesNuxt agent-readiness guidelines for making a site operable by autonomous AI agents — not just cited by them. Covers the isitagentready.com standards: Markdown content negotiation, RFC 8288 Link headers, RFC 9727 API catalogs, Agent Skills discovery indexes, WebMCP browser tools,
SKILL.md
| 1 | # Nuxt Agent-Ready Best Practices |
| 2 | |
| 3 | Guidelines for making a Nuxt 4 site **operable by autonomous AI agents** — measured by the [isitagentready.com](https://isitagentready.com) scanner (Cloudflare's "Is Your Site Agent-Ready?"). This is a different axis from GEO: GEO is about being _cited_ in AI answers; agent-readiness is about being _operated_ — an agent authenticating, discovering your API, calling your tools, and taking action. |
| 4 | |
| 5 | Proven on production Nuxt 4 + Nitro `cloudflare_module` Workers: |
| 6 | |
| 7 | - A **marketing site** (only public POST endpoints, no auth/MCP server): **21 → 50+ (Level 1 → Level 4 "Agent-Integrated")** — the auth + MCP surfaces are honesty-gated OFF (see below). |
| 8 | - A **full platform** with a real OAuth server (Better-Auth oauth-provider) + a real remote MCP server: **21 → 100/100 (Level 5 "Agent-Native"), all 14 checks green**. The auth + MCP surfaces are legitimately publishable there, which is what unlocks the last ~40 points. |
| 9 | |
| 10 | **The ceiling is set by what you actually run, not by effort.** A marketing site tops out around Level 4 and that is the _correct_ score — do not fabricate an auth server to chase 100 (see THE HONESTY RULE). Only a site with a real authorization server and a real MCP server can honestly reach Level 5. |
| 11 | |
| 12 | ## GEO vs Agent-Readiness (know the difference) |
| 13 | |
| 14 | | | GEO (`nuxt-geo-best-practices`) | Agent-Readiness (this skill) | |
| 15 | | -------- | ---------------------------------------------------------- | ---------------------------------------------------------------------- | |
| 16 | | Goal | Be **cited** in AI answers | Be **operated** by agents | |
| 17 | | Question | "Will ChatGPT mention me?" | "Can an agent call my tools and act?" | |
| 18 | | Levers | llms.txt, crawler allowlist, RAG content, JSON-LD entities | MCP/WebMCP, API/skill discovery, agent auth, DNS-AID, agentic commerce | |
| 19 | |
| 20 | **Shared primitives live in the GEO skill.** `robots.txt` AI-crawler allowlisting, `llms.txt`/`llms-full.txt`, and the XML sitemap are covered by `nuxt-geo-best-practices` (rules `ai-robots-allowlist`, `ai-llms-txt`, `ai-sitemap`). The isitagentready scanner scores those too — set them up via the GEO skill first, then apply this skill for the agent-operation layer on top. |
| 21 | |
| 22 | ## When to Apply |
| 23 | |
| 24 | - Raising a site's score on `https://isitagentready.com/<domain>` |
| 25 | - Supporting `Accept: text/markdown` content negotiation for agents |
| 26 | - Advertising resources via RFC 8288 `Link` headers |
| 27 | - Publishing an RFC 9727 API catalog (`/.well-known/api-catalog`) |
| 28 | - Publishing an Agent Skills discovery index (`/.well-known/agent-skills/index.json`) |
| 29 | - Exposing site actions to in-browser agents via WebMCP (`navigator.modelContext`) |
| 30 | - Publishing an MCP Server Card (only if you run a real MCP server) |
| 31 | - Publishing OAuth/OIDC discovery for agent auth (only if you run a real auth server) |
| 32 | - Publishing DNS-AID records (`_index._agents.<domain>`) + DNSSEC |
| 33 | |
| 34 | ## THE HONESTY RULE (load-bearing — read before publishing anything) |
| 35 | |
| 36 | **Only publish discovery for services that actually exist.** A discovery document that sends an agent to a dead end — a `/.well-known/openid-configuration` with no auth server behind it, an MCP Server Card whose transport endpoint 404s, a `_mcp._agents` DNS record with no MCP server — is **worse than a lower score**. Agents will try to use it and fail. |
| 37 | |
| 38 | This mirrors the "never fake customers/logos/scale" rule: a fabricated capability that fails on first contact destroys trust. On a marketing site with only public POST endpoints, **skip** OAuth/OIDC discovery, oauth-protected-resource, auth.md, and the MCP Server Card — they belong on the app/console domain (real auth server) or require building a real MCP server. Decline these explicitly and say why. |
| 39 | |
| 40 | **The flip side — when you DO run the real thing, publish it fully.** If your site runs a real OAuth authorization server (e.g. Better-Auth `oauth-provider` plugin) and a real remote MCP server, the auth + MCP surfaces are no longer dishonest — they are the highest-value checks and unlock Level 5. maps.guru scored 100/100 precisely because those services exist. The honesty rule cuts both ways: don't fake it, but don' |
People who liked this also viewed
6 picksType
- find-skillsHelps users discover and install agent skills when they ask questions like "how do I do X", "find a skill for X", "is there a skill that can...", or express interest in extending capabilities. This skSkillsJul 2026
- grill-meA relentless interview to sharpen a plan or design.SkillsJul 2026
- vercel-react-best-practicesVercel's official collection of agent skillsSkillsJul 2026
- improve-codebase-architectureScan a codebase for deepening opportunities, present them as a visual HTML report, then grill through whichever one you pick.SkillsJul 2026
- setup-matt-pocock-skillsConfigure this repo for the engineering skills — set up its issue tracker, triage label vocabulary, and domain doc layout. Run once before first use of the other engineering skills.SkillsJul 2026
- triageMove issues and external PRs through a state machine of triage roles — categorise, verify, grill if needed, and write agent-ready briefs.SkillsJul 2026
