Python: Add search tool content for OpenAI responses

[eavanvalkenburg]

Motivation and Context

OpenAI Responses web and file search built-in tools currently show up in raw provider output, but the Python OpenAI chat client drops them from the framework content model. This change preserves those built-in tool calls/results so downstream surfaces can observe them the same way they already can for code interpreter, shell, image generation, and MCP tool activity.

Description

• add search_tool_call and search_tool_result content types in core
• map OpenAI non-streaming web_search_call and file_search_call output items into those content types
• map OpenAI streaming response.output_item.added / response.output_item.done search items into those content types and explicitly ignore search progress events instead of leaving them as unparsed noise
• forward seen_reasoning_delta_item_ids through both OpenAI streaming paths so reasoning done events are suppressed after deltas consistently
• add unit tests covering the new search parsing and the streamed reasoning-dedup paths

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the Contribution Guidelines
• All unit tests pass, and I have added new tests where possible
• Is this a breaking change? If yes, add "[BREAKING]" prefix to the title of the PR.

[moonbox3]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/core/agent_framework
_types.py	1092	87	92%	58, 67–68, 122, 127, 146, 148, 152, 156, 158, 160, 162, 180, 184, 210, 232, 237, 242, 246, 276, 689–690, 849–850, 1285, 1357, 1392, 1412, 1422, 1474, 1606–1608, 1790, 1893–1898, 1923, 2011, 2019–2021, 2026, 2117, 2129, 2152, 2407, 2431, 2526, 2775, 2982, 3055, 3066, 3068–3072, 3074, 3077–3085, 3095, 3165, 3300, 3305, 3310, 3315, 3319, 3403–3405, 3434, 3511–3515
packages/openai/agent_framework_openai
_chat_client.py	914	121	86%	522–525, 529–530, 536–537, 572–578, 599, 607, 630, 748, 847, 906, 908, 910, 912, 978, 992, 1072, 1082, 1087, 1130, 1246, 1427, 1432, 1436–1438, 1442–1443, 1509, 1538, 1544, 1554, 1560, 1565, 1571, 1576–1577, 1596, 1686, 1708–1709, 1724–1725, 1743–1744, 1787, 1953, 1991–1992, 2008, 2010, 2089–2097, 2127, 2234, 2269, 2284, 2304–2314, 2327, 2338–2342, 2356, 2370–2381, 2390, 2422–2425, 2435–2436, 2447–2449, 2463–2465, 2475–2476, 2482, 2497
TOTAL	27730	3199	88%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
5615	20 💤	0 ❌	0 🔥	1m 31s ⏱️

[Copilot]

Pull request overview

This PR updates the Python OpenAI Responses chat client and the core content model to preserve OpenAI built-in web/file search tool activity (calls/results) in the framework’s unified Content representation, including streaming support and reasoning-done de-duplication.

Changes:

• Add new unified content types: search_tool_call and search_tool_result in core.
• Parse OpenAI non-streaming web_search_call / file_search_call output items into those new content types.
• Parse OpenAI streaming search output items (response.output_item.added / response.output_item.done), explicitly ignore search progress events, and forward seen_reasoning_delta_item_ids consistently across both streaming paths.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
python/packages/openai/agent_framework_openai/_chat_client.py	Maps Responses API search output items into unified Content objects; ignores search progress events; forwards reasoning delta tracking through both streaming paths.
python/packages/core/agent_framework/_types.py	Extends ContentType and adds Content.from_search_tool_call() / Content.from_search_tool_result() constructors; updates argument parsing docstring.
python/packages/openai/tests/openai/test_openai_chat_client.py	Adds unit tests for non-streaming + streaming search parsing and for reasoning done suppression after deltas across both streaming implementations.

Comments suppressed due to low confidence (1)

python/packages/core/agent_framework/_types.py:1543

• Content.parse_arguments() docstring states it raises ContentError when the content type doesn't support arguments, but the implementation doesn't check self.type and will never raise that error. Please either add an explicit type guard (e.g., allow only function_call/mcp_server_tool_call/search_tool_call) or update the docstring to reflect the actual behavior.

    def parse_arguments(self) -> dict[str, Any | None] | None:
        """Parse arguments from function_call, mcp_server_tool_call, or search_tool_call content.

        If arguments cannot be parsed as JSON or the result is not a dict,
        they are returned as a dictionary with a single key "raw".

        Returns:
            Parsed arguments as a dictionary, or None if no arguments.

        Raises:
            ContentError: If the content type doesn't support arguments.