Docs

How AgentShelf works, step by step

And exactly where its boundary sits.

How it works

The delivery pipeline

Send your store URL

We run the AgentShelf checker (the same public endpoints an agent uses) + a human-grade product-language review.

Scored report in 72 hours

Your score, every failed check with evidence, and fixes ranked by impact.

Fix Pack writes the fixes

Merchant-specific agents.md, llms.txt, rewritten titles/descriptions for your top products, variant labels, FAQ + shipping/returns in agent-readable plain text — paste-ready, nothing changed on your store without you.

The 9 checks, as specimens

The checker reads the same public endpoints an agent uses: agents.md llms.txt /.well-known/ucp robots.txt /products.json — then grades the catalog itself: title clarity, description substance, variant pricing and compare-at traps, and variant labels.

01agents.md present & merchant-specific12 points

02llms.txt present8 points

03UCP discovery (/.well-known/ucp)12 points

04robots.txt allows AI agents10 points

05Machine-readable catalog (/products.json)14 points

06Product titles agent-clear (>=12 chars)10 points

07Descriptions substantive (>=200 chars text)12 points

08Every variant priced, no compare-at traps12 points

09Variants machine-labeled (no 'Default Title')10 points

Verdicts

Score 80 or above: AGENT-READY. 50–79: AGENT-VISIBLE, BUY-BUTTON RISK. Under 50: AGENT-INVISIBLE / HIGH RISK. Without a readable catalog, only checks 1–5 can run — a maximum of 56 of 100 points. Non-Shopify stores use the SOP feed fallback, labeled in the report and never presented as endpoint-verified.

Sentinel

Monthly re-run; diff against the last score; any regression (new thin titles, compare-at trap, template regression, robots/UCP change) produces a drift report and fix list within 5 business days.

FAQ

Which platforms?

Shopify first-class (public catalog endpoints). WooCommerce/Etsy via feed + page review fallback in the SOP — same scorecard, catalog check adapted.

Do you change my store?

Never directly. Fix Pack delivers paste-ready files/copy; you (or your dev) apply them. Re-check after applying is included in Fix Pack.

What do you check?

The 9 checks above + product-language review. Full method in delivery/agentshelf_check.py — it's the actual tool we run.

Guarantees?

Score/finding accuracy against the live store at audit time, and delivery in 72 hours or the audit is free. We do not guarantee agent placement, traffic, or sales.

Is this legal?

Yes — we only read public endpoints/pages any agent or browser can read. No scraping behind logins, no credential access.

Your next customer is a robot.

Agents don’t browse — they read your product data, your agents.md, your policies, and your compare-at prices, then decide. Find out what they see.

Get my AgentShelf Audit — $149 See pricing