You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
adcp 5.7.0 declares a buyer_ref field on the media-buy success response types, but buyer_ref was removed from the AdCP spec (create/update media-buy requests and responses) in the 3.1 line. Since the SDK types are generated from the spec, these fields should not be present.
Affected types (5.7.0)
CreateMediaBuySuccessResponse — has buyer_ref
CreateMediaBuyResponse1 (union arm) — has buyer_ref
UpdateMediaBuySuccessResponse — has buyer_ref
UpdateMediaBuyResponse1 (union arm) — has buyer_ref
(The corresponding request types — CreateMediaBuyRequest, UpdateMediaBuyRequest — correctly have nobuyer_ref, which is consistent with the spec. The divergence is response-side only.)
Spec evidence
In the pinned spec schemas (commit 04f59d2d5 of adcontextprotocol/adcp, the 3.1 derivation), neither create-media-buy-response.json nor update-media-buy-response.json declares buyer_ref anywhere in their success arms. The success arms expose media_buy_id, status, revision, implementation_date, etc. — no buyer_ref. buyer_ref appears only in prose (error-recovery suggestions, account-authorization field-scope examples), never as a response property.
Impact
Downstream servers that build responses from these SDK models will emit a non-spec buyer_ref field on the wire (when extra/serialization passes it through), and conformance tooling that grades responses against the 3.1 response schemas will see an unexpected field.
Expected
Regenerate the success-response types from the 3.1 spec so buyer_ref is dropped from CreateMediaBuySuccessResponse / UpdateMediaBuySuccessResponse (and their union arms).
Repro
fromadcp.typesimportUpdateMediaBuySuccessResponse, CreateMediaBuySuccessResponseassert"buyer_ref"notinUpdateMediaBuySuccessResponse.model_fields# fails on 5.7.0assert"buyer_ref"notinCreateMediaBuySuccessResponse.model_fields# fails on 5.7.0
code-reviewer: On main (v6.3.0b10), both underlying classes are already clean — CreateMediaBuyResponse1 and UpdateMediaBuyResponse1 carry no buyer_ref field. The four names in the issue reduce to two classes: CreateMediaBuySuccessResponse and UpdateMediaBuySuccessResponse are type aliases defined in aliases.py:504 and aliases.py:719.
ad-tech-protocol-expert: Spec evidence is sound — AdCP 3.1 response schemas contain no buyer_ref property; its presence on success-response types is a straightforward spec-drift bug, not a deliberate extension.
My take: The fix landed on main as part of the same codegen pass that closed #914 and covered both the Create and Update response variants, though the Update half wasn't explicitly documented in that issue. The blocker is release: #946 is open and tracking whether a v5.7.1 patch is feasible, but the prior triage there flagged the cherry-pick as non-trivial (the codegen diff bundles additional constraint tightenings) and escalated the v5.x maintenance-track decision to @bokelley.
Blocked-on: #946 — resurfaces on merge or on a maintainer decision about the v5.x maintenance track.
Resolved on the current stable line by the regenerated AdCP 3.1.x types, now released in 6.6.0. The remaining question was the v5.7.x backport tracked in #946; closing this with the GA release path available.
Activity
fix(schemas): drop SDK-inherited buyer_ref from media-buy success res…
bokelley commentedon Jun 12, 2026
Triage
Classification: Bug
Bucket(s): client
Status: deferred
What the experts said:
main(v6.3.0b10), both underlying classes are already clean —CreateMediaBuyResponse1andUpdateMediaBuyResponse1carry nobuyer_reffield. The four names in the issue reduce to two classes:CreateMediaBuySuccessResponseandUpdateMediaBuySuccessResponseare type aliases defined inaliases.py:504andaliases.py:719.buyer_refproperty; its presence on success-response types is a straightforward spec-drift bug, not a deliberate extension.My take: The fix landed on
mainas part of the same codegen pass that closed #914 and covered both the Create and Update response variants, though the Update half wasn't explicitly documented in that issue. The blocker is release: #946 is open and tracking whether a v5.7.1 patch is feasible, but the prior triage there flagged the cherry-pick as non-trivial (the codegen diff bundles additional constraint tightenings) and escalated the v5.x maintenance-track decision to @bokelley.Blocked-on: #946 — resurfaces on merge or on a maintainer decision about the v5.x maintenance track.
Triaged by Claude Code. Session: https://claude.ai/code/${CLAUDE_CODE_REMOTE_SESSION_ID}
Generated by Claude Code
bokelley commentedon Jul 1, 2026
Resolved on the current stable line by the regenerated AdCP 3.1.x types, now released in 6.6.0. The remaining question was the v5.7.x backport tracked in #946; closing this with the GA release path available.
test: transport-aware BDD harness — retire e2e_rest ledger 312→7 (#1418…