Skip to content
← Field notes

Field notes

What is Agentrix

In one line: Agentrix runs a real AI agent against your site and measures whether it can get through.

Published guideagentic-commerce

What it measures

Agentrix drives a real browser with an AI agent (Claude's computer-use) through signup → login → browse → cart → pre-checkout, recording which steps it reached, where it stopped, and why. That becomes a 0-100 Agentrix Score, computed the same way every time — not a gut call.

You can compare a run against the previous one (/reports/[id]/compare) to see whether things are actually improving over time.

What it doesn't measure

Agentrix is not a substitute for a security audit, penetration test, vulnerability assessment, or accessibility audit. It has a different purpose than tools that statically scan config or code.

AgentrixConfig checkers & vulnerability scanners
What it measuresCan agents reach signup → pre-checkout?Config, code, known vulns, etc.
How it runsA real agent walks a real browserScans and rule matching
PurchaseNever clicks purchase confirmOut of scope (or separate process)
DeliverableScore, why it stopped, screenshotsFinding lists and severity labels

Safety design

  • No confirmer name or date, no diagnosis (consent gate)
  • The harness mechanically refuses purchase/payment confirmation clicks — a purchase is never completed
  • Stops immediately on CAPTCHA/bot defenses — never attempts a bypass
  • Auto-stops past 40 actions or $2
  • Identifies as a dedicated bot (AgentQA-Bot/1.0) — never impersonates a regular browser

See Safety design and the Glossary for details. For the reasoning, read Why Agentrix now; for the mechanics, How Agentrix works.

Try it on your own site

Paste a URL and see right away: whether AI crawlers are let in, whether there is a file written for AI, and whether the page is marked to stay out of search — using public pages only.