Skip to main content
Skill test-sales-agent · version 1.0.0 · for sellers.

​
When to use

  • A seller is connecting its own AdCP sales agent, or its ad server through a managed sales agent.
  • An inventory source shows an error or unhealthy state and the seller wants to check it again.
  • A seller wants to see how its storefront agent answers a brief before real buyers send one.

​
Before you start

  • You are connected to the seller account with a role that can change its storefront.
  • You know the agent’s endpoint URL, its protocol (MCP or A2A) and its authentication type.
  • Any credential is already in your environment. Never ask the person to paste it into chat.

​
Steps

  1. Call get_status and confirm the seller account.
  2. Call open_seller_setup with focus: "inventory_sources" to show the setup checks.
  3. Call search with kind: "inventory_source". If the source already exists, keep its id and update it instead of adding a second one.
  4. Call save_inventory_source with name, executionType (AGENT for your own sales agent), endpointUrl, protocol, authenticationType, credentials, productMode and runDiscoveryTest: true. Pass sourceId when updating. Confirm with the person if asked.
  5. Call get with kind: "inventory_source" and the id. Read its health and the discovery test result. If the test failed, report the error, fix the URL or authentication with the person, and run step 4 again.
  6. Send a practice brief. Call save_rfp with action: "create", origin: "manual", purpose: "evaluation", a request (brief, budget, currency, flight, channels, countries) and a clientRequestId.
  7. Call open_proposal_pass with the rfpId and turnId, or get with kind: "rfp". Check that the products came from the source you connected and that prices respect your floors.
  8. To coach the storefront agent, call save_rfp with action: "append_turn", the parentTurnId and a new request.instruction. Record your verdict with action: "record_feedback" and a grade.
  9. Call open_seller_setup with focus: "publish_validation" to see whether the end-to-end sandbox check has passed.

​
Guardrails

  • Practice briefs use purpose: "evaluation". Never create a live RFP to test.
  • Credentials are write-only; never echo them, log them or ask for them in chat.
  • Do not change the listing, marketplace participation, playbook or AI Business Rules as part of a test unless the person asks.
  • A test never contacts real buyers and never books real spend.

​
Done when

  • The inventory source reports healthy and its discovery test returned products.
  • A practice RFP turn shows products from that source.
  • You have given the person the evidence: source id, health, RFP and turn ids, the products returned, and every issue with its suggested fix.