Track rc4 codegen follow-ups from Argus review

[bokelley]

Follow-ups from Argus review on PR #906. These were non-blocking for the rc4 schema update, but should be tracked before the next schema regen that introduces named open-payload classes.

• Extend scripts/post_generate_fixes.py _set_class_extra_allow to handle compact model_config = ConfigDict(extra=...) forms and add a unit test for compact extra='forbid' -> extra='allow'.
• Harden _ensure_configdict_import for files without an existing from pydantic import ... line.
• Rework _open_payload_class_names title-less root fallback so it cannot target the first enum in a multi-class generated file.
• Revisit tests/test_extra_policy.py::_schema_allows_extra handling of x-adcp-open-payload: false; either leave it unhandled until the spec lands or fail CI if source schemas set it.
• Document the rc4 verify_brand_claim(s) success response change where signed_response is now required.
• Consider stabilizing or documenting PricingCurrency, which is codegen-derived from an inline schema.
• Improve SCHEMA_DELTAS.md output for human-readable rc.2 -> rc.4 changes instead of reporting only "No field-shape changes detected."

Source: PR #906 Argus review, submitted 2026-05-30.

[bokelley]

Triage

Classification: Enhancement — codegen script hardening + rc4 docs follow-ups (tracking issue)
Bucket(s): client (items 1–3), docs (item 5)
Status: ready-to-implement

What the experts said:

• code-reviewer: Items 1–3 are latent correctness bugs — compact ConfigDict falls through to "insert fresh block" producing two model_config declarations; missing pydantic import leaves ConfigDict undefined in the generated file; first-^class match for the title-less root fallback silently targets enum classes in multi-class files; all non-breaking, one PR; item 4 blocked-on feat(codegen): consume x-adcp-open-payload in generated models #902.
• docs-expert: Item 5 is clean additive docs (non-breaking, types already enforce signed_response required); item 6 needs a design call first (alias vs. dedup vs. prose, plus two duplicate PricingCurrency definitions to resolve); item 7 is a codegen/CI snapshot-management problem, not a prose edit.

My take: Items 1–3 are correctness-blocking for the next schema regen that introduces named open-payload classes or compact ConfigDict forms — they should land before that regen. Items 6–7 are pre-design; they should remain open as follow-ons with a decision from @bokelley on scope.

Items 1–3 — one PR (scripts/post_generate_fixes.py + tests/test_extra_policy.py):

• _set_class_extra_allow: add compact model_config = ConfigDict(extra=...) one-liner match before the multi-line fallback; a second model_config block breaks Pydantic at import time
• _ensure_configdict_import: when ConfigDict is in content but no from pydantic import line exists, prepend from pydantic import ConfigDict instead of silently returning
• _first_generated_class_name / _open_payload_class_names: exclude classes whose header contains an Enum base from the title-less root sentinel so multi-class files are safe
• Unit tests for compact extra='forbid' → extra='allow', no-pydantic-import injection, multi-class file with leading enum; no public API changes, no semver bump

Item 4: Blocked-on #902 — the x-adcp-open-payload: false enforcement decision belongs inside that work. Add an inline comment in _schema_allows_extra marking the passive short-circuit as intentional pending #902.

Item 5 (separate PR): Add CHANGELOG entry and a docs note for signed_response now required on VerifyBrandClaimSuccessResponse; the generated types already enforce this, docs is the only gap.

Items 6–7: Deferred — item 6 needs a design decision (which of alias/dedup/prose, and which of the two duplicate PricingCurrency definitions is authoritative); item 7 needs snapshot-management architecture in generate_types.py/CI. Both should stay open here as named follow-ons.

---

Triaged by Claude Code. Session: https://claude.ai/code/cse_01SwCsRM6poDDTB97iiT7aJd

---

Generated by Claude Code