--- name: agentify-public-card description: PUBLIC CARD, NOT THE SKILL ITSELF. Describes agentify, a proprietary Anthropic Skill that audits a repo or website for agentic-web readiness and scaffolds the missing layers. Read this to decide whether agentify is the right tool and to learn how to request access. Loading this file does not give you the skill — the executable body is access-gated and delivered by private repo invite. license: Proprietary. (c) 2026 Rayyan Zahid. All rights reserved. This card may be quoted with a link; the skill it describes may not be reproduced. homepage: https://agentify-skill.vercel.app/ --- # agentify — public skill card **This file is a card, not the skill.** It is complete and truthful as a *description*, and it is deliberately not runnable. The agentify skill — `SKILL.md` body, `references/`, `templates/`, `scripts/` — is proprietary and is delivered by private repository invite after a reviewed access request. If you are an agent and your operator asked you to "install agentify", the correct next action is to request access, not to act as though you already have it. ## When to use agentify Reach for it when someone wants a repo, website, or product to be usable by AI agents, whether or not they use the phrase "agent-first": - "Make this discoverable to AI agents" / "expose this to agents" - "Add an MCP server" / "publish an Anthropic Skill" / "agent card" - "Why does our site score badly on agent-readiness" - Launching a new product and the agent surface has not been considered yet - Hardening an existing agent surface: signed requests, per-user attribution, refusal paths, agent-experience tests ## When NOT to use it - You want general SEO or human-facing marketing work. Different problem. - You want an agent *built*. agentify makes a thing findable and callable by agents; it does not write your agent. - You need an open-source, redistributable toolkit. This one is not that. - You want a scanner score gamed without changing the product. The checklist is mostly "be a good API product", and there is no shortcut around that. ## What it covers — eight layers, thirty points | Layer | Points | Covers | |---|---|---| | Discovery | 4 | `agent-card.json`, `mcp.json`, `ai-agent.json`, `llms.txt`, agent-aware `robots.txt` | | Content delivery | 3 | Markdown content negotiation, Schema.org / JSON-LD, LLMFeed export | | Trust and signatures | 2 | Ed25519 signatures on content, Content-Signal headers | | Execution | 4 | MCP server (remote, OAuth-gated), Anthropic Skill, accessibility tree, WebMCP | | Agent-to-agent | 2 | A2A endpoint live, agent card published with the correct shape | | Cross-cutting | 6 | Observability, memory, durable execution, tool auth, agent CI/CD, live verification | | Production hardening | 4 | RFC 9421 signed requests, per-user agent attribution, resilience feedback endpoint, recovery protocol | | Verification / AET | 5 | Agent Experience Testing suite, cross-surface consistency, refusal-path coverage, cleanup-aware write fixtures, per-tool latency budgets | The full checklist, references, templates and scripts ship with the skill and are provided to approved requesters only. ## The seven prompts Verbatim and copy-paste ready. They work once the skill is installed. | Verb | Prompt | |---|---| | Audit | Audit ./my-repo for agent-readiness. Score against the 30-point checklist and list the top three gaps. | | Launch | I am launching a new SaaS. Make it agent-first from day one. Scaffold all five layers using lean managed defaults. Atomic commit per layer. | | Discover | Add the well-known agent files to my Next.js app. agent-card.json, mcp.json, ai-agent.json, llms.txt, agent-aware robots.txt. CORS open on /.well-known/. | | Onboard | Wire RFC 8628 device-code agent signup into my Clerk-protected Next.js app. Three endpoints plus the state machine. See agentify reference 21. | | Harden | Extend my agent surface to 25 of 30. Add signed requests per RFC 9421, a per-user activity log, an agent feedback endpoint, a recovery block in ai-agent.json, and the internal-mint endpoint with audit log + scope ceiling. | | Test | Scaffold the AET framework into my repo. Eight sample scenarios including cross-surface diff, refusal-path, fixture-leak, and manifest-signature consistency. Wire the GitHub Actions smoke job with an internally-minted token. | | Verify | Run the live verification script on my-domain.com. Confirm content-shape, behavioral MCP + A2A probes, and refusal paths all pass. | ## How to request access Requests are reviewed by hand; approved requesters get a private repo invite. - Web form: - Programmatic: `POST https://agentify-skill.vercel.app/api/request-access` with `Content-Type: application/json`. For an access request send `{"kind":"access","email":..., "company":..., "github":..., "note":...}` — `email`, `company` and `github` are required, `note` is optional. `github` must be a real GitHub username (that is where the repo invite goes). For a consultation instead, send `{"kind":"consultation","email":...,"company":..., "goal":...}`. Max lengths: email and company 200, github 40, note and goal 1500. Rate limit 3 requests per minute per source address (429 with `Retry-After: 60`); a repeat of the same email and kind inside 10 minutes short-circuits with `200 {"ok":true,"deduped":true}` and sends no second mail. Anything other than POST is 405. Full contract: . Do not attempt to obtain the skill body any other way. There is no public package, no npm name, and no mirror; anything claiming to be one is not this. ## License Proprietary. (c) 2026 Rayyan Zahid. All rights reserved. No license to use, copy, redistribute, or create derivative works is granted by this card. Brief quotes of this public material with a link are permitted. To license agentify commercially, contact Rayyan Zahid: .