v2.5 brand_manifest adapter preserves URL path, breaks brand.domain regex

[bokelley]

Repro

Send a v2.5 get_products request with a brand_manifest URL that includes a path:

client.call_tool("get_products", {"brand_manifest": "https://acme.com/.well-known/brand.json", "brief": "..."})

Observed

INVALID_REQUEST[brand.domain]: get_products failed:
Invalid value for field 'brand.domain':
String should match pattern '^[a-z0-9]([a-z0-9-]*[a-z0-9])?(\.[a-z0-9]([a-z0-9-]*[a-z0-9])?)*$'

Root cause

In adcp/compat/legacy/v2_5/get_products.py:130-133:

# brand_manifest (v2.5 URL string) → brand.domain (v3 BrandReference)
brand_manifest = out.pop("brand_manifest", None)
if isinstance(brand_manifest, str) and brand_manifest and "brand" not in out:
    out["brand"] = {"domain": strip_url_scheme(brand_manifest)}

strip_url_scheme (in adcp/compat/legacy/v2_5/_url.py) only strips the scheme prefix and trailing slash. For https://acme.com/.well-known/brand.json it returns acme.com/.well-known/brand.json — which has slashes and dots in positions that fail BrandReference.domain's regex.

The SDK's own _normalize_brand_manifest in adcp/server/translate.py:485 (the utility intended for adopters to call manually) gets this right — it uses urlparse(manifest).hostname. Same concept, different impl, only one of the two extracts the hostname correctly.

Spec context

brand_manifest per the v2.5 spec is documented as "https://example.com/brand.json" — i.e. a URL to a JSON file, which by construction has a path. So the legitimate input shape is exactly the one this adapter breaks.

Impact

Surfaced via shape-based legacy detection (#673) in 5.2.0 — any v2.5 buyer (untagged or otherwise) sending a brand_manifest URL with a path gets a confusing brand.domain regex error. Trapped us on a CI run where the test had been passing under 5.1.0 (where brand_manifest got silently stripped as an unknown field, no translation attempted).

Suggested fix

from urllib.parse import urlparse

if isinstance(brand_manifest, str) and brand_manifest and "brand" not in out:
    parsed = urlparse(brand_manifest)
    domain = parsed.hostname or strip_url_scheme(brand_manifest)
    out["brand"] = {"domain": domain}

Same shape as _normalize_brand_manifest in translate.py. The strip_url_scheme fallback covers the case where the buyer sent a bare domain without :// (no scheme → hostname is None).

Also applies to adapt_request in create_media_buy.py and update_media_buy.py if they take the same path (haven't checked).

Environment

• adcp==5.2.0
• v2.5 wire-version buyer (or shape-detected via is_legacy_shape)

[bokelley]

Triage

Classification: Bug
Bucket(s): client
Status: drafting-pr
Milestone: (omitted — no target version named)

What the experts said:

• code-reviewer: approved — urlparse().hostname is the correct primitive; docstring accurately covers IPv6/port/bare-domain behaviour; regression tests present at both call sites; no blockers
• dx-expert: approved — helper-in-_url.py is the right DRY approach; both call sites updated consistently; nits on strip_url_scheme public naming and IPv6 silent failure are non-blocking follow-ups

My take: Bug confirmed in both get_products.adapt_request and _media_buy_helpers.adapt_brand_manifest_to_brand (used by create_media_buy). update_media_buy carries no brand_manifest and is unaffected. Fix is non-breaking — adds extract_brand_domain() using urlparse().hostname with a bare-domain fallback; swaps both call sites. Draft PR with regression tests: #679.

---

Triaged by Claude Code. Session: https://claude.ai/code/session_01YX14H75zYWQ1BYuopyUL34

---

Generated by Claude Code