Python: fix build_model_schema putting pydantic constraint objects into field descriptions

[Sanjay Santhanam (Sanjays2402)]

The bug

KernelJsonSchemaBuilder.build_model_schema() pulled the field description from FieldInfo.metadata[0] when metadata was a non-empty list. In pydantic v2, metadata is the list of constraint objects (Ge, Le, ...), so a field like

top_p: float = Field(default=0.9, ge=0.0, le=1.0, description="nucleus sampling")

got description=Ge(ge=0.0) in the generated schema. The schema then failed json.dumps(...) with TypeError: Object of type Ge is not JSON serializable, and the real description was silently dropped.

Reproduced on current main:

from pydantic import BaseModel, Field
import json
from semantic_kernel.schema.kernel_json_schema_builder import KernelJsonSchemaBuilder

class Params(BaseModel):
    top_p: float = Field(default=0.9, ge=0.0, le=1.0, description="nucleus sampling")

json.dumps(KernelJsonSchemaBuilder.build_model_schema(Params)["properties"]["top_p"])
# TypeError: Object of type Ge is not JSON serializable

The fix

python/semantic_kernel/schema/kernel_json_schema_builder.py: use field_info.description (the documented pydantic attribute) instead of field_info.metadata[0]. The dict-metadata fallback is kept for non-pydantic field-info objects.

Tests

• Added test_build_model_schema_with_field_constraints_keeps_description to python/tests/unit/schema/test_schema_builder.py — asserts the real description survives and the schema is JSON-serializable. Verified it fails on the unfixed code and passes with the fix.
• Ran the schema builder unit suite: 37 passed.
• ruff check and ruff format --check are clean on both touched files.

Fixes #14443.

[Copilot]

🔵 Needs a closer look

Restore string metadata fallback for Annotated field descriptions while excluding constraint objects.

Pull request overview

Fixes Pydantic constraint objects being emitted as non-serializable field descriptions.

Changes:

• Uses field_info.description for Pydantic fields.
• Adds regression coverage for constrained fields and JSON serialization.

File summaries

File	Changes
python/tests/unit/schema/test_schema_builder.py	Adds regression test coverage.
python/semantic_kernel/schema/kernel_json_schema_builder.py	Updates field-description extraction.

Review details

Suppressed comments (1)

python/semantic_kernel/schema/kernel_json_schema_builder.py:98

• This removes the list-metadata fallback wholesale, but list metadata is also how Pydantic v2 stores arbitrary Annotated metadata. The existing Items model in this file uses Annotated[str, "Description of the item"]; its FieldInfo.description is None, so this change silently drops that nested field description. Keep a fallback only for string metadata, and only when .description is absent/None, so Ge/Le objects remain excluded while Annotated descriptions continue to work.

                elif hasattr(field_info, "description"):
                    # Pydantic v2 keeps constraints (Ge, Le, ...) in field_info.metadata,
                    # which is a list of constraint objects and not a description,
                    # so the description must come from field_info.description instead.
                    field_description = field_info.description

• Files reviewed: 2/2 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.