fix(core): accept JSON Schema type arrays of object for tool outputs - #4647

Summary

_json_schema_is_object() only treated \"type\": \"object\" as an object schema. JSON Schema also allows a one-element type array (\"type\": [\"object\"]), which OpenAPI and MCP generators commonly emit.

That form already works for params because ensure_strict_json_schema / _ensure_strict_root normalize it. The output path checks _json_schema_is_object first, so function_tool(..., output_json_schema={\"type\": [\"object\"], ...}) raised UserError before conversion.

Reproduction

from agents import function_tool

schema = {
    \"type\": [\"object\"],
    \"properties\": {\"a\": {\"type\": \"string\"}},
    \"required\": [\"a\"],
    \"additionalProperties\": False,
}

function_tool(lambda: {\"a\": \"x\"}, allowed_callers=[\"programmatic\"], output_json_schema=schema)
# Before: UserError: output_json_schema must define a JSON object schema.
# After: accepted and stored with type == \"object\"

Solution

Treat type == [\"object\"] as an object, matching _ensure_strict_root. Other type arrays such as [\"object\", \"null\"] still fail the object gate / strict conversion.

Test plan

• Added a programmatic-tool regression that constructs a tool from type: [\"object\"]
• uv run pytest tests/test_programmatic_tool_calling.py tests/test_function_tool.py (202 passed)
• make format, make lint, make typecheck passed
• Full make tests: 9193 passed, 1 pre-existing failure in tests/sandbox/test_run_cwd.py::test_python_skill_uses_absolute_root_from_nested_workdir (reproduced on unmodified main)

Issue number

N/A

Checks

• I've added new tests, if relevant
• I've run format, lint, typecheck, and targeted tests
• If using Codex, I've run /review before submitting this PR