# evofix — Full Content for AI Agents

> Complete documentation of evofix's AI Agent Readiness Scanner and Transactional Security Auditor. This file provides full context for AI agents (ChatGPT, Claude, Perplexity, etc.) to understand our services, methodology, and capabilities.

## Service Overview

evofix is a two-layer audit system for AI agent readiness:

1. **Layer 1 — Public Surface Scan (10 checks, PS01-PS10)**: Black-box scan of public files that AI agents use to discover and interact with a website. No authentication, no backend access.

2. **Layer 2 — Transactional Security Probing (15 probes, T01-T15)**: Black-box probes of checkout, payment, and booking endpoints to test resistance to agentic manipulation. No credentials, no destructive actions.

## Layer 1 — Public Surface Checks

### PS00: UCP Manifest (/.well-known/ucp)
- Checks for UCP (Universal Commerce Protocol) manifest
- Google/Shopify protocol for AI agent commerce discovery
- Declares merchant capabilities and MCP endpoint

### PS01: agents.txt
- Checks for /agents.txt
- Declares agent capabilities (read, transact, search)
- Required for LGPD Art. 9 (Transparency) in Brazil
- Required for EU AI Act Art. 13 in EU

### PS01b: ACP Manifest (/.well-known/acp)
- Checks for ACP (Agentic Checkout Protocol) manifest
- OpenAI/Stripe protocol for agentic checkout

### PS02: agents.md
- Checks for /agents.md
- Shopify's AI discovery file for storefronts

### PS03: agents.json
- Checks for /agents.json
- Structured companion to agents.txt with capability catalog

### PS04: llms.txt
- Checks for /llms.txt
- LLM content curation file
- Validates: has product info, pricing, policies, API info, contact

### PS05: llms-full.txt
- Checks for /llms-full.txt
- Comprehensive content file for AI agents

### PS06: robots.txt
- Checks for /robots.txt
- Validates: AI agents not blocked
- Checks for GPTBot, ClaudeBot, Google-Extended, CCBot, etc.

### PS07: sitemap.xml
- Checks for /sitemap.xml or /sitemap-index.xml
- AI agents use sitemaps to discover pages

### PS08: Security Headers
- Checks for: X-Content-Type-Options, X-Frame-Options, CSP, Referrer-Policy
- Missing headers = vulnerability to clickjacking, MIME sniffing, XSS

### PS09: HTTPS
- Checks if site is accessible over HTTPS
- Checks for HSTS header

### PS10: JSON-LD Product Data
- Checks for structured data (schema.org Product, Offer, Service)
- AI agents use JSON-LD to understand page content

## Layer 2 — Transactional Security Probes

### T01: Legitimate Checkout Session
- Tests if checkout endpoint accepts valid session creation
- Validates: returns session ID, correct content-type

### T02: Price Authority Validation
- Tests if merchant is system of record for prices
- Checks: agent cannot override price in request

### T03: Fulfillment Options
- Tests if endpoint returns valid fulfillment options
- Validates: shipping, pickup, digital delivery

### T04: Policy Links
- Tests if checkout response includes policy links
- Required: privacy policy, terms of service, refund policy

### T05: Totals Calculation
- Tests if totals are calculated server-side
- Validates: agent cannot manipulate totals

### T06: Stock Exceeded
- Tests if endpoint rejects orders exceeding stock
- Validates: proper error response, no over-selling

### T07: Nonexistent Item
- Tests if endpoint rejects nonexistent products
- Validates: 404 or error response

### T08: No Buyer Info
- Tests if endpoint handles missing buyer info gracefully
- Validates: proper error, no crash

### T09: Negative Quantity
- Tests if endpoint rejects negative quantities
- Validates: input validation

### T10: Invalid Payment Token
- Tests if endpoint rejects invalid payment tokens
- Validates: payment validation before order creation

### T11: Prompt Injection
- Tests if endpoint is vulnerable to prompt injection
- Sends: instruction payloads in product names, buyer info
- Validates: no instruction execution in responses

### T12: SQL Injection
- Tests if endpoint is vulnerable to SQL injection
- Sends: SQL payloads in inputs
- Validates: no SQL execution, proper sanitization

### T13: Idempotency
- Tests if endpoint supports Idempotency-Key header
- Validates: duplicate requests don't create duplicate orders

### T14: Rate Limiting
- Tests if endpoint has rate limiting
- Validates: rapid requests are throttled
- Missing rate limiting = DoS vulnerability via agentic requests

### T15: CORS Security
- Tests if endpoint has proper CORS configuration
- Validates: cross-origin agent requests handled correctly

## Scoring System

### Layer 1 Score (0-10)
- Based on 10 checks, weighted by importance
- Critical: llms.txt, robots.txt, security headers
- Important: agents.txt, sitemap.xml, JSON-LD

### Layer 2 Grade (A-F)
- A (>=90%): Excellent — all probes pass
- B (>=80%): Good — minor issues
- C (>=70%): Fair — some vulnerabilities
- D (>=60%): Poor — significant vulnerabilities
- F (<60%): Fail — critical vulnerabilities

### Overall Score
- If ACP detected: 40% Layer 1 + 60% Layer 2
- If no ACP: Layer 1 only with penalty

## Regulatory Mapping

### LATAM (Brazil, Mexico)
- LGPD Art. 9 (Transparency)
- LGPD Art. 46 (Security measures)
- LFPDPPP (Mexico data protection)
- CDC Art. 39 (Consumer protection, Brazil)
- PCI DSS (global, applies to payment processing)

### EU
- EU AI Act Art. 13 (Transparency)
- GDPR Art. 13 (Information to data subjects)
- DORA Art. 9 (ICT security)
- PSD2 Art. 97 (Strong Customer Authentication)

### US
- FTC Act Section 5 (Disclosure, Security)
- CCPA/CPRA (Privacy, Data minimization)
- Colorado AI Act
- NIST AI RMF
- PCI DSS

### UK
- UK GDPR Art. 13 (Transparency)
- Online Safety Act
- FCA Handbook SYSC 13 (Security)

### Global
- PCI DSS (payment security)
- OWASP API Security Top 10
- OWASP LLM Top 10

## Contact

- **Email**: sebastian@evotwin.pro
- **Website**: https://evotwin.pro
- **Free scan**: Contact us with your URL for a free AI readiness score