[bokelley]

Optional maintenance fallback: adcontextprotocol/adcp#8636 selects this SDK's next branch for 3.3 schema CI, where both SDK generation jobs now pass. This stable-line backport is retained as a validated alternative and is not required for that integration path.

The spec repository's current default-branch checkout cannot generate the composed 3.3 schema tree because the repairs in #1705 and #1741 landed on next.

Backport both repairs: resolve inlined core references at every caller depth, retain composed envelope ancestry and narrowed fields, and locate generated classes by name. The response-emitter conflict preserves main's existing explicit-null handling. The SDK stays pinned to 3.2.7 and its generated output remains byte-identical.

Validation in Docker:

• Stable 3.2.7 bundle, Python 3.11: 156 focused generator, hierarchy, class-matcher and schema-pin tests pass; make validate-generated confirms byte-identical output.
• Current 3.3 source bundle, Python 3.12: full make regenerate-schemas, make validate-generated, and the same 156 tests pass. This includes export consolidation, envelope ancestry, narrowed fields and version-field retention.
• The new composed-envelope regression reproduces the missing-reference failure against the original main generator.
• Normal pre-commit and commit-message hooks pass, including formatting, lint, mypy and adopter type checks.

The candidate run mirrors spec CI's disposable pin/manifest adjustment and candidate-only sync isolation. It uses the deterministic bundle built at adcontextprotocol/adcp commit 95e1447549daf96ea4f29b9a879f7fe4090a4a59, including #8547, with SHA-256 d3a612e34b0e6ddb533f7d5c867ddccca9697f4734f743bfde0f2fba924876c7. This validates the current source bundle; the existing published beta.2 artifact remains unchanged.

Refs #1680 and adcontextprotocol/adcp#8547. The implementation is backported from #1705 and #1741.

[aao-secretariat]

Ladon verdict: Approve

Approve — codegen-tooling backport (scripts + tests only).

Checked:

• No findings from the reviewer (critical/high/medium/low all clear).
• Diff touches only scripts/generate_types.py, scripts/post_generate_fixes.py, and three test files — no public adcp.* surface, no wire-shape change, generated output claimed byte-identical for the 3.2.7 pin.
• New pin_conflicting_allof_properties logic is fail-closed (raises on undecidable oneOf/anyOf-branch disagreement rather than guessing), and stabilize_inlined_core_refs rewrites resolve correctly via existing _normalize_schema_ref_target.
• find_class_header/require_class_header and normalize_emitted_envelope_bases correctly follow generated class composition; strong new test coverage.
• fix(codegen) semver signal is correct for a non-breaking tooling change.

Gate checks: no no-auto-approve team match; gated_paths false; high_risk false. No blocking or medium findings → decision table falls through to row 9 (approve). Note: review_decision is REVIEW_REQUIRED, but with gated_paths: false row 2 does not apply.