Most audits end the same way: a report lands, someone is supposed to read it, prioritise it, and turn it into tickets. For agent-readiness that’s doubly ironic — the audit says AI agents can’t buy from your store, and then hands a human a to-do list.

There’s a faster loop now, because the person fixing a store increasingly isn’t a person. Merchants and their developers work with AI coding agents — Claude Code, Cursor and their peers — that can edit a theme, add a JSON-LD block, publish a .well-known file and open a pull request. What those agents need isn’t a dashboard. They need one file that contains the whole job.

That’s what the agent brief is.

What’s in the brief

GET /api/v1/scan/<scan_id>/brief.md returns a single markdown document, generated deterministically from the scan’s own parsed data — no model wrote it, so nothing in it is hallucinated. It contains:

  • The verdict and the plan — the store’s score, and the sequenced steps ranked by points recovered (“41 → 68, grade D → B”), each with its exact how-to and the check ID it closes, so success is verifiable per step.
  • Seven paste-ready artifacts, built from the store’s own catalog: a Product JSON-LD block, an llms.txt, a robots.txt block allowing the AI agent user-agents, an A2A Agent Card and an ARD AI Catalog for /.well-known/, an attribute mapping table, and — for Shopify stores — the JSON-LD fix as a Liquid snippet that covers every product page at once.
  • The verification loop — the exact API call to re-scan the store after deploying, and what the response must show (the listed checks flipping to pass, the score moving toward the projection).

The two .well-known files are worth singling out: publishing an A2A Agent Card and an ARD AI Catalog are the two easiest checks in the audit, almost nobody on the web serves them yet, and the brief ships both files pre-filled. Two pastes, six points.

The whole flow, honestly

  1. Scan your store — free. The scan itself, the score, what’s wrong in plain language, plus your llms.txt and the robots.txt block are free.
  2. The brief embeds the fix how-tos and the catalog-specific artifacts, which are the paid part. Pro comes with an API key (Dashboard → Account → API key).
  3. Give your agent the key and one instruction. Something like:

Fetch https://www.agentready.market/api/v1/scan/<scan_id>/brief.md with the header Authorization: Bearer <your key>. Apply every step to my store’s codebase, then re-run the scan as the brief describes and report the before/after score.

  1. The agent applies the steps, re-scans through the same API, and the score movement is the receipt. Nothing in the loop requires trusting anyone’s claim — the re-scan is the proof.

The same file works fine for a human developer; markdown reads both ways. The point isn’t that an agent must do it — it’s that the deliverable no longer assumes a human will.

Why the file is deterministic

A brief written by a language model could invent a fix, a field, or a price. This one can’t: every value is either a fact the scan parsed from the store (product names, currencies, options, which files answered) or an explicit <placeholder> the merchant must fill. That discipline — the same one behind the score itself — matters more, not less, when the consumer is an agent that will apply the file without skepticism.

Docs

The API surface is documented for machines first: /api.md, OpenAPI 3.1, and the discovery document. The brief endpoint requires a Pro Bearer key and returns text/markdown; without a key it returns a structured 401 that points at the free (masked) JSON report instead.

AgentReady at agentready.market, operated by Boxing Vision — unaffiliated with other AgentReady-named products. The scan measures whether an AI agent can find, understand and buy on a store; the free scan is the place to start.