Training agent: serve brand.json format from get_brand_identity

[bokelley]

Context

The training agent's get_brand_identity handler returns a custom BrandEntry format (flat object with brand_id, house object, names, industries, etc.). Real seller agents serve the actual brand.json schema — $schema, house (domain string), brands[] array with identity fields like name, voice, logos, colors, fonts.

The storyboard compliance tests in @adcp/client validate against the brand.json spec. Returning a non-standard shape means the training agent's responses don't match what a real agent would return.

What needs to happen

1. Refactor handleGetBrandIdentity to return brand.json format directly
2. For sandbox brands loaded via getSandboxBrands(), serve brand_json as-is (scoped by fields/authorization)
3. For talent brands (Loti Entertainment), restructure seed data to match brand.json format
4. Update tests to assert brand.json shape

Related

• chore: release package adcp-client#515 (type brand_json with Zod schema)
• feat: add governance, content standards, and creative approval tools to test agent #2135 (added governance/content tools to training agent)

[bokelley]

Re-verified 2026-04-16 against https://test-agent.adcontextprotocol.org/mcp/ — STILL BROKEN

get_brand_identity still returns the flat BrandEntry shape:

{
  "brand_id": "acme_outdoor",
  "house": { "domain": "acmeoutdoor.example", "name": "Acme Outdoor" },
  "names": [{"en": "Acme Outdoor"}],
  "industries": ["retail"],
  "keller_type": "master",
  "logos": [...],
  "tagline": "",
  "available_fields": ["colors", "fonts", "tone"]
}

Not the brand.json well-known shape (no $schema, no brands[] array, house is an object rather than a domain string). The brand-protocol get-brand-identity-response.json schema passes because the agent is validating against that schema — but it's a different shape than the well-known brand.json file a real seller would publish.