Python: feat: enable dynamic tool loading by passing tools list in kwargs [progressive exposure]

[suneetnangia]

This pull request enhances the agent framework to support dynamic tool loading, enabling tools to add new capabilities during execution. The main changes ensure that the current tools list is passed to tool functions, allowing them to register additional tools on the fly, this allows progressive tools exposure for LLM instead of loading all tools upfront. A new sample is also provided to demonstrate this feature in action.

Framework improvements for dynamic tool loading:

• Updated function_invocation_wrapper and streaming_function_invocation_wrapper to pass the current tools list in the kwargs to tool functions, allowing tools to dynamically add more tools during execution. [1] [2]

Testing and sample updates:

• Added assertions in the test mock_get_response to verify that the tools list is present in kwargs and is a list, ensuring correct propagation for dynamic tool loading.
• Added a new sample script dynamic_tool_loading.py demonstrating how a tool can add new tools (e.g., factorial and fibonacci calculators) to the agent at runtime, and how these tools can be used immediately within the same run.

[markwallace-microsoft]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/core/agent_framework
_tools.py	774	74	90%	226, 272, 323, 325, 353, 523, 555–556, 658, 660, 680, 698, 712, 724, 729, 731, 738, 771, 827–829, 870, 895–904, 910–919, 955, 965, 1206, 1543, 1630–1634, 1773, 1775, 1841, 1935, 1941, 1985–1986, 1999–2000, 2043, 2129, 2167–2168, 2209–2211, 2249–2250, 2260, 2317–2318, 2325–2326
TOTAL	17033	2562	84%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
3317	213 💤	0 ❌	0 🔥	1m 4s ⏱️

[Copilot]

Pull request overview

Enables progressive/dynamic tool exposure by forwarding the active tools list to tool functions via **kwargs, allowing tools to register additional tools during execution.

Changes:

• Pass the current tools collection into tool runtime kwargs in both non-streaming and streaming function invocation wrappers.
• Extend kwargs-propagation tests to assert the tools list is present in tool **kwargs.
• Add a new Python sample demonstrating dynamic tool loading at runtime.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 7 comments.

File	Description
python/packages/core/agent_framework/_tools.py	Injects the current tools list into custom_args passed to tool execution for dynamic tool registration.
python/packages/core/tests/core/test_kwargs_propagation_to_ai_function.py	Adds assertions ensuring tools is propagated in tool **kwargs.
python/samples/getting_started/tools/dynamic_tool_loading.py	New sample showing a tool that adds factorial/fibonacci tools during the same agent run.

[Copilot]

In this sample, if not tools_list: will treat an empty list as missing. Since the framework can legitimately pass an empty tools list, this should distinguish None/missing from an empty list (and ideally validate the type before branching) so the error path isn’t triggered incorrectly.

Suggested change

	if not tools_list:
	if tools_list is None:

[suneetnangia]

Not sure if the suggestion is valid in this instance.
Since we are checking in the tool itself, framework should never pass is an empty i.e. there should be at least 1 tool?

[suneetnangia]

Closing in favour of #3877 @markwallace-microsoft, @eavanvalkenburg