This package contains a reusable skill for creating an Agent Readiness Layer around a website, business, product, API, or documentation system.
It helps turn a normal human-facing site into an agent-operable business surface: a layer that AI agents can discover, understand, trust, compare, recommend, and use safely.
Version 3 adds Agent Discovery Hardening: the machine-readable layer is not considered complete until a cold agent can find it without being explicitly told that /llms.txt, /llms-full.txt, or /docs/ exist.
Version 4 adds the Broadcast Protocol: a standardized set of signals that make the machine layer discoverable through nine independent paths — so any agent, regardless of how it was built, can find it.
- LLM-friendly documentation
llms.txtandllms-full.txtAGENTS.md— dedicated agent entry point with start-here chain- Markdown mirrors and source-of-truth docs
- Agent discovery and recommendation profiles
- Agent discovery hardening plans with broadcast signal checklist
/docs/HTML index (no JavaScript required)/sitemap.md,robots.txthints, alternate markdown links, redirect aliases/.well-known/ai-plugin.json— cross-platform plugin discovery- Content negotiation endpoint for
Accept: text/markdown - Static fallbacks for JavaScript-rendered pages
- Machine-readable page profiles
- Schema.org JSON-LD and metadata maps
- OpenAPI/MCP readiness maps
- Tool/action definitions
- Agent action matrices
- Permission and identity models
- Human approval and escalation rules
- Memory policies
- Observability and audit requirements
- Sandbox and evaluation scenarios
- Agent journey maps
This skill is not just SEO, GEO, AEO, or README generation.
It is AXO: Agent Experience Optimization.
SEO asks: can a search engine rank this?
AXO asks:
- Can an AI agent discover this business?
- Can a cold agent find the machine-readable layer without being told the file paths?
- Can it understand what the business does?
- Can it compare the offer against alternatives?
- Can it recommend the correct next step?
- Can it use the right tool or API safely?
- Can it avoid hallucinated claims?
- Can the business audit what happened afterward?
The skill uses eight layers:
- Eyes — structure and machine readability
- Discovery — how agents find, classify, compare, and reach the machine-readable layer
- Context — source-of-truth docs, markdown mirrors,
llms.txt, and retrieval surfaces - Hands — APIs, forms, tools, MCP servers, and actions
- Permits — identity, scopes, approvals, safety, and auditability
- Brain — intent logic, decision rules, conversion routing, and refusal boundaries
- Memory — durable context, brand facts, user preference boundaries, and privacy
- Evaluation — scenario tests, sandbox tests, hallucination traps, and regression checks
A machine-readable layer is incomplete until it is discoverable from at least five independent paths. Target nine.
| Signal | Who reads it |
|---|---|
<link rel="alternate" type="text/markdown" href="/llms.txt" title="AI Agent Docs"> in <head> |
HTML-inspection agents, head-parsers |
Hidden anchor <a href="/llms.txt" style="display:none"> as first child of <body> |
Sequential DOM-parsers, scrapers that skip <head> |
<section role="doc-instructions" aria-labelledby> in footer |
Browser agents, computer-use agents navigating by ARIA landmarks |
robots.txt with Link: </llms.txt>; rel="help" + AI docs comment block |
Crawlers that parse robots.txt for agent hints |
sitemap.xml with all machine files listed |
Agents that start from robots → sitemap |
/.well-known/ai-plugin.json with description_for_model pointing to docs |
OpenAI-compatible agents, agents that check .well-known |
/docs/ as a no-JavaScript HTML index |
Agents following links without JS execution |
/AGENTS.md with numbered start-here chain |
Coding agents, GitHub-pattern agents |
/api/negotiate with Accept: text/markdown → returns llms.txt |
Agents that use HTTP content negotiation |
Each signal is independent. An agent that misses one can find the machine layer through another.
Start with SKILL.md. Then use the templates in /templates, checklists in /checklists, snippets in /snippets, and evals in /evals to generate a full Agent Readiness Layer for a specific website or business.
For planning-only work, use the framework and checklists without generating files.
For implementation work, generate the requested artifacts using the templates.
This skill is designed to align with the modern agent infrastructure ecosystem:
- Agent Skills /
SKILL.mdpackage format llms.txt,llms-full.txt,AGENTS.md, markdown mirrors, andsitemap.md- Schema.org JSON-LD
- OpenAPI / Swagger tool definitions
- MCP-style tool/resource/prompt surfaces
.well-known/ai-plugin.json(OpenAI plugin standard)- DPUB ARIA roles for browser-agent navigation
- HTTP content negotiation (
Accept: text/markdown) - Agent identity, least-privilege scopes, and audit logs
- Browser-agent and computer-use readiness
- Sandbox testing and deterministic evals
- Agent discoverability and recommendation optimization
/public
llms.txt
llms-full.txt
AGENTS.md
sitemap.xml
sitemap.md
robots.txt
/.well-known
ai-plugin.json
/api
negotiate.js (content negotiation serverless function)
/docs
index.html (no-JS HTML docs index)
index.md (markdown mirror)
business-profile.md
source-of-truth.md
agent-discovery-profile.md
agent-journey-map.md
agent-discovery-hardening-plan.md
context-retrieval-map.md
services.md
pricing.md
locations.md
faq.md
offers.md
target-audience.md
brand-voice.md
conversion-rules.md
agent-behavior-rules.md
permission-model.md
agent-identity-model.md
memory-policy.md
tool-readiness-map.md
mcp-readiness-map.md
openapi-action-definitions.md
browser-agent-readiness.md
agent-action-matrix.md
observability-audit-plan.md
evaluation-plan.md
missing-info-report.md
/schema
organization.json
local-business.json
website.json
webpage.json
services.json
faq.json
breadcrumbs.json
/metadata
page-metadata-map.md
alt-text-map.md
/evals
agent-readiness-evals.md
sandbox-scenarios.md
cold-agent-crawl.md
README.md