feat: expose output schema to model input filters

[its-amann]

Summary

• Exposes the effective output_schema on ModelInputData for call_model_input_filter.
• Lets filters replace the schema for streamed and non-streamed model calls, including final-output parsing after get_new_response.
• Preserves the existing schema when filters only edit input or instructions, and keeps ToolOutputTrimmer from dropping schema metadata.
• Documents switching a structured-output call back to plain text with AgentOutputSchema(str).

Closes #3563

Tests

• uv run ruff check src/agents/run_config.py src/agents/run_internal/turn_preparation.py src/agents/run_internal/run_loop.py src/agents/extensions/tool_output_trimmer.py tests/test_call_model_input_filter_unit.py tests/extensions/test_tool_output_trimmer.py tests/test_agent_runner.py tests/test_server_conversation_tracker.py
• uv run ruff format --check src/agents/run_config.py src/agents/run_internal/turn_preparation.py src/agents/run_internal/run_loop.py src/agents/extensions/tool_output_trimmer.py tests/test_call_model_input_filter_unit.py tests/extensions/test_tool_output_trimmer.py tests/test_agent_runner.py tests/test_server_conversation_tracker.py
• uv run pyright --project pyrightconfig.json src/agents/run_config.py src/agents/run_internal/turn_preparation.py src/agents/run_internal/run_loop.py src/agents/extensions/tool_output_trimmer.py tests/test_call_model_input_filter_unit.py tests/extensions/test_tool_output_trimmer.py tests/test_agent_runner.py tests/test_server_conversation_tracker.py
• uv run mypy src/agents/run_config.py src/agents/run_internal/turn_preparation.py src/agents/run_internal/run_loop.py src/agents/extensions/tool_output_trimmer.py tests/test_call_model_input_filter_unit.py tests/extensions/test_tool_output_trimmer.py tests/test_agent_runner.py tests/test_server_conversation_tracker.py --exclude site
• uv run pytest tests/test_call_model_input_filter_unit.py tests/extensions/test_tool_output_trimmer.py tests/test_agent_runner.py::test_conversation_lock_rewind_skips_when_no_snapshot tests/test_agent_runner.py::test_get_new_response_uses_agent_retry_settings tests/test_server_conversation_tracker.py::test_get_new_response_marks_filtered_input_as_sent tests/test_server_conversation_tracker.py::test_run_single_turn_streamed_marks_filtered_input_as_sent -q

[its-amann]

@codex review

[chatgpt-codex-connector]

Note

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[seratch]

Thanks for the contribution. The implementation is thoughtful, but the linked issue does not currently demonstrate a concrete use case that requires changing the output schema inside call_model_input_filter. Most known cases can select Agent.output_type before the run or use a separately configured agent.