Fixed scope
$250
One-time review of one merchant or platform origin. Delivery target: 7 days after scope acceptance.
Merchants, platforms, and agent operators
Decide which doors AI agents may use—before traffic finds them.
A fixed-scope evidence audit that converts a vague allow-or-block posture into an action-level access map for discovery, comparison, cart, checkout, and support.
Access ladder
- Public discovery and comparison facts.
- Verified agent identity and declared purpose.
- Authenticated cart, account, and support scopes.
- Buyer-confirmed checkout and charge-bearing actions.
Evidence controls
- Robots, terms, agent instructions, and API policies agree.
- The agent and represented customer remain distinct, including whether agent origin is verified or inferred.
- Customer delegation never substitutes for merchant permission.
- Inbound triggers, communications consent, opt-outs, and long-running follow-up remain linked.
- Discovery URLs, catalog/card integrity, and action authority remain distinct.
- Handoff, purchase, support, return, claim, reversal, and attribution remain linked.
Deliverable
- Route-by-route access matrix and exact public evidence.
- Conflicts, unknowns, and current denial behavior.
- Prioritized remediation and retest criteria.
- One bounded public-surface retest.
Why now. The Bank for International Settlements distinguishes open agentic-commerce systems from closed systems where a retailer can require its own agent and block others. Read the October 5, 2026 speech.
Crossmint's production-only Agent Checkouts API can traverse protocol-enabled merchants or an automated browser, use merchant-supported payment methods, apply a hard spending cap, pause for authorization, and return a receipt. This makes route-specific merchant permission, buyer authority, and outcome evidence operational requirements—not assumptions.
Riskified's October 8, 2026 announcement separates an agent credential from the trustworthiness of the represented customer and carries verified-or-inferred agent provenance into orders, refunds, claims, and support decisions. Mastercard's Agent Pay Trust Framework independently combines agent identity, represented-human context, proof of intent, authority limits, and behavioral risk signals.
inride's October 9, 2026 Trade Agent Acquire launch shows an inbound valuation form triggering immediate text or email and automated nurturing over days, weeks, or months. Its separate outbound product advertises number validation, consent checks, and opt-out suppression. These are provider claims, but they expose a practical evidence boundary between a form submission, a permitted channel, the duration and scope of follow-up, and the final CRM handoff.
The Linux Foundation-hosted AI Catalog working repository defines a draft discovery envelope for heterogeneous AI artifacts. Its current text permits signatures and trust manifests, but a catalog signature is not authenticity without a separate signer policy. Formal A2A and MCP adoption is still pending, so catalog presence and integrity are recorded without treating either as permission or certification.
The audit does not assume open access is always correct; it makes each permitted and prohibited action explicit and testable. Provider claims are evidence inputs, not certification or proof of production performance.