Agent policy
Agent-initiated purchases
Autonomous agents may search the catalogue and create orders on this site. Accepted
authorisation: an AP2 mandate signed with the user's passkey, or an ACP shared payment token
issued for this merchant. Card numbers typed by an agent are never accepted.
Liability
An order created with a valid mandate is binding on the user who signed it; the agent operator
is responsible for staying inside scope_amount_max. Orders without a valid mandate
are cancelled and refunded in full.
Spending limits
Default ceiling 2,000 USD per order and 10,000 USD per 30 days; write to
agents@shop.example.com to raise it.
Refunds and shipping
14-day free returns; see /returns for the machine-readable policy.
```
Note: The agent word and the purchase word must appear in the visible text of the same page, so a heading-only mention or a bare link to a PDF does not register.
Verify with:
```bash
DOMAIN=shop.example.com
for p in /ai-policy /agent-policy /terms; do printf '%s: ' "$p"; curl -sL "https://$DOMAIN$p" | sed 's/<[^>]*>/ /g' | grep -Eio 'agent|purchase|order|checkout|refund' | sort -u | tr '\n' ' '; echo; done
```
---
### 14. AGENTS.md with the required sections — `agents-md`
Status: **fail** · Tier: recommended · Impact: high · Dimension: Discoverability
Measured: "Not found (HTTP 404, 11026 bytes)."
Observed:
```text
HTTP 404 · text/html · 11026 bytes · 48 ms · final https://www.shopify.com/AGENTS.md
```
Why it matters: The instruction file for coding and operations agents. It fails more often than any other check in the curated corpus we measured, and the usual reason is not a missing file but a file without the sections an agent needs.
Change to make:
```markdown
# AGENTS.md
## Project overview
Example Inc is a payments API for marketplaces. Docs: https://example.com/docs
## Setup and build
- `pnpm install`
- `pnpm dev` -> http://localhost:3000
- `pnpm build`
## Testing
- `pnpm test`; lint and types: `pnpm lint && pnpm typecheck`
## Conventions
- TypeScript strict mode; no `any` in exported signatures.
- Public API routes live in `src/routes/api/`.
## Security and contact
- Never commit secrets; use `.env.local`.
- Report vulnerabilities to security@example.com.
```
Note: AGENTS.md is plain markdown with no mandated fields, so the headings above are the sections this audit looks for - and in a monorepo the file nearest the edited code wins.
Verify with:
```bash
DOMAIN=example.com
curl -sS "https://$DOMAIN/AGENTS.md" | head -30
```
---
### 15. Content-Signal usage signals — `content-signal`
Status: **fail** · Tier: recommended · Impact: high · Dimension: Discoverability
Measured: "robots.txt answered HTTP 429 and the homepage response carries no Content-Signal header."
Observed:
```text
HTTP 429 · text/html; charset=utf-8 · 9358 bytes · 29 ms
```
Why it matters: Declares how your content may be used, inside robots.txt. Cloudflare included this in its readiness checks.
Change to make:
```text
# https://example.com/robots.txt - Content Signals Policy (contentsignals.org)
# search = index the content and return links and short excerpts
# ai-input = use the content in a model at answer time (grounding)
# ai-train = use the content to train or fine-tune a model
User-agent: *
Content-Signal: search=yes, ai-input=yes, ai-train=no
Allow: /
User-agent: GPTBot
Content-Signal: search=yes, ai-input=yes, ai-train=no
Allow: /
```
Note: The directive is only read inside a user-agent group, values are case-insensitive yes or no, and omitting a signal neither grants nor restricts that use.
Verify with:
```bash
DOMAIN=example.com
curl -sS "https://$DOMAIN/robots.txt" | grep -i -B1 '^content-signal:'
```
---
### 16. FAQ / question-answer schema — `faq-schema`
Status: **fail** · Tier: recommended · Impact: high · Dimension: Content Accessibility
Measured: "No FAQPage entity among 1 parsed JSON-LD block on the homepage."
Observed:
```text
@types: Corporation, ContactPoint
```
Why it matters: Question-and-answer content is the type agents quote most.
Change to make:
```html
```
Note: The same questions must be visible as text on the page — structured data that does not match rendered content is treated as spam.
Verify with:
```bash
DOMAIN="https://example.com"
curl -sL "$DOMAIN/" | perl -0777 -ne 'while (m{