fix(agent_tool_input): summarize Pydantic Optional[T] anyOf fields

[adityasingh2400]

Summary

• The schema summary helper rejected any field with an anyOf key, but Pydantic renders T | None as anyOf: [{type: T}, {type: "null"}]. Because _summarize_json_schema returns None for the whole schema when any single field is unsupported, one Optional[T] field would silently suppress the entire structured-input summary, including sibling fields with valid types and descriptions.
• Detect the 2-branch anyOf of a simple type plus null and render it as T | null, matching the existing type: ["T", "null"] path.
• Other anyOf shapes (unions of two non-null types, anyOf with nested objects, etc.) are still rejected.

Repro

from pydantic import BaseModel, TypeAdapter
from agents.agent_tool_input import build_structured_input_schema_info

class Args(BaseModel):
    count: int | None = None  # Pydantic emits anyOf: [{type:integer},{type:null}]
    name: str  # has a description in the rendered schema

schema = TypeAdapter(Args).json_schema()
# add descriptions for the demo
schema["properties"]["count"]["description"] = "Optional count."
schema["properties"]["name"]["description"] = "A name."

print(build_structured_input_schema_info(schema, include_json_schema=False).summary)
# Before: None  (entire summary dropped)
# After:
#   - count (integer | null, optional) - Optional count.
#   - name (string, required) - A name.

Test plan

• New test test_schema_summary_handles_pydantic_optional_anyof covers the nullable anyOf shape and confirms non-nullable anyOf unions are still rejected.
• Test fails on main (assert None is not None) and passes with the fix.
• All existing tests/test_agent_tool_input.py, tests/test_agent_tool_state.py, and tests/test_agent_as_tool.py tests still pass (60/60).

[seratch]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d6ae9a108b

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Preserve enum/const constraints in nullable anyOf

Pydantic emits Optional[Literal[...]] as anyOf with an enum/const branch that also has type: "string". This helper now accepts that branch as a bare simple type, so the schema summary says e.g. string | null instead of the allowed literals, misleading the nested agent about valid structured input.

Useful? React with 👍 / 👎.

[adityasingh2400]

Addressed in c22eb566: Preserve enum/const branches in the nullable anyOf summary so Optional[Literal[...]] reports the allowed literals instead of bare string | null. Test added for the literal case.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 64e23d389c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Guard non-string anyOf branch types

When a supported parameter model supplies a JSON Schema anyOf branch whose type uses the valid list form, for example via Pydantic json_schema_extra as {"type": ["integer", "string"]} plus a null branch, this membership test hashes the list and raises TypeError while building the agent tool. Other unsupported schema shapes in this helper fall back to None, so this should first verify entry_type is a string before checking _SIMPLE_JSON_SCHEMA_TYPES.

Useful? React with 👍 / 👎.