Regenerate models from ucp@c5c6139 (Cart, Catalog, identity-linking, signed totals)

[caohy1988]

What

Regenerates the SDK against Universal-Commerce-Protocol/ucp@c5c61396ebf9a5afb9c4169bfa3af1409daa6fd6 — the identity-linking OAuth 2.0 foundation merge from PR #354 plus all v2026-04-08 features. Adds the small generator/preprocessor changes that the regen depends on, plus semantic smoke tests so this regression mode is caught next time.

Opening as draft to give maintainers a chance to weigh in on (a) the --ref flag shape and (b) whether you'd prefer to wait for the next dated release/* cut before landing identity-linking models. Happy to rebase on whichever ref you prefer.

Why

Downstream observability work (Universal-Commerce-Protocol-Analytics issue #8) needs typed access to:

• Cart (POST /carts, GET /carts/{id}, …) — was missing from the previous snapshot
• Catalog Search + Lookup, including GetProductRequest / GetProductResponse / DetailProduct from catalog_lookup.json#/$defs
• common/identity_linking (config.scopes, scope_policy) introduced by ucp PR #354
• All four Context fields — intent, language, currency, eligibility (the previous snapshot exposed intent only)
• Signed totals — Total.amount is now SignedAmount (a signed integer) so refunds and discounts can carry negative values without losing the sign
• Open total / error / info / warning vocabularies — the spec documents these as well-known values that businesses MAY extend, so the generated types must be str / RootModel[str], not closed Literal[...]
• Standalone info_code / warning_code models alongside error_code

release/2026-04-08 is not sufficient because identity-linking, info_code, and warning_code merged after the release branch was cut.

[gemini-code-assist]

Code Review

This pull request updates the UCP SDK by regenerating schema models to support new capabilities including Cart, Catalog, and identity linking, alongside a transition to signed totals. It also introduces provenance tracking for generated models and adds semantic smoke tests to prevent regressions. Feedback identifies a potential failure point in the generation script where an invalid git reference would not halt execution, potentially leading to models being generated from the wrong version.

I am having trouble creating individual review comments. Click here to see my feedback.

generate_models.sh (50)

The git checkout command is executed in a subshell without checking its exit status. If the provided ref is invalid, the script will silently continue using the default branch of the cloned repository, resulting in models generated from the wrong version while recording the intended ref in provenance. You should check the exit status or rely on set -e.

(cd ucp && git checkout "$REF_VALUE") || { echo "Error: Failed to checkout ref $REF_VALUE" >&2; exit 1; }

[caohy1988]

Blocking review

I checked PR 1a against issue #8 acceptance and ran the PR tests locally:

uv run pytest -q
9 passed

I found two blockers before this should move out of draft.

1. generate_models.sh --ref can silently generate from the wrong ref.

   generate_models.sh does not use set -euo pipefail, and the arbitrary-ref path does:

   git clone https://github.com/Universal-Commerce-Protocol/ucp ucp
   (cd ucp && git checkout "$REF_VALUE")
   UCP_SCHEMA_REF="$(cd ucp && git rev-parse HEAD)"

   If git checkout "$REF_VALUE" fails, the script keeps going on the clone's default branch and records that HEAD as provenance. That undermines the main reason issue chore: add CODEOWNERS file #8 asked for explicit-ref/SHA support.

   Fix: add strict shell mode and/or explicitly check every critical command:

   set -euo pipefail
   ...
   git clone https://github.com/Universal-Commerce-Protocol/ucp ucp
   git -C ucp checkout "$REF_VALUE"

   Also add a small generator test or shell smoke check proving ./generate_models.sh --ref definitely-not-a-real-ref exits nonzero.

2. identity_linking is importable but not semantically generated.

   Issue chore: add CODEOWNERS file #8 acceptance requires common/identity_linking.py with config.scopes / scope_policy. This PR's generated module is only:

   • src/ucp_sdk/models/schemas/common/identity_linking/__init__.py: class IdentityLinking(RootModel[Any])
   • src/ucp_sdk/models/schemas/common/identity_linking/dev/ucp/common.py: also RootModel[Any]

   ScopePolicy and ScopeToken exist, but there is no typed business identity-linking model exposing config.scopes. The current test only imports IdentityLinking, so it passes even though downstream cannot use typed access for the field issue chore: add CODEOWNERS file #8 explicitly called out.

   Fix options:

   • adjust preprocessing/codegen so the business schema under $defs["dev.ucp.common.identity_linking"].business_schema produces a typed class with config.scopes, or
   • if the nested reverse-domain $defs layout prevents that, add a targeted generated/helper model for this schema and cover it with a semantic test.

   Acceptance test should instantiate/access something like:

   model = IdentityLinkingBusinessSchema(
       config={"scopes": {"dev.ucp.shopping.order:read": {}}}
   )
   assert "dev.ucp.shopping.order:read" in model.config.scopes

Everything else I checked aligns with PR 1a's current scope: Cart/Catalog files are present, Context has intent/language/currency/eligibility, totals use signed amounts with open type, message codes are open RootModel[str], and provenance records c5c61396ebf9a5afb9c4169bfa3af1409daa6fd6.