Python: Fixed declarative deep research sample

[dmytrostruk]

Motivation and Context

• Created missing workflow.yaml implementing the Magentic orchestration pattern with dynamic agent dispatch
• Fixed _get_agent_name() in InvokeAzureAgentExecutor to evaluate PowerFx expressions for dynamic agent names
• Fixed JoinExecutor to initialize workflow state when used as the entry node, so agents receive the user input
• Fixed _get_branch_exit() in DeclarativeWorkflowBuilder to skip terminator actions (GotoAction, EndWorkflow, etc.), preventing dual-edge wiring that caused infinite loops
• Simplified path resolution in main.py
• Added 8 unit tests covering expression-based agent name resolution and terminator branch exit handling
• Fix also applies to declarative/marketing sample.

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.

[markwallace-microsoft]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/declarative/agent_framework_declarative/_workflows
_declarative_builder.py	345	24	93%	176, 233, 324, 346, 356, 407, 409, 417, 432, 468, 471, 497–498, 517–518, 594, 599, 643–645, 669–670, 892, 965
_executors_agents.py	479	86	82%	94–95, 112–114, 155–156, 163, 191, 194–195, 203–204, 214, 225–230, 509, 547–549, 552, 568, 584–588, 612, 617, 707, 713–718, 725, 749–750, 759–761, 810–811, 819, 838, 945–948, 955–956, 963–964, 976–981, 985–986, 988, 993, 995, 999–1002, 1005, 1007–1008, 1011, 1019–1021, 1023, 1026–1027, 1029, 1056–1057
_executors_control_flow.py	170	6	96%	81, 89, 152, 342–343, 347
TOTAL	21259	3303	84%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
4198	239 💤	0 ❌	0 🔥	1m 13s ⏱️

[Copilot]

Pull request overview

This PR fixes and completes the Python declarative “Deep Research” workflow sample and corrects several declarative workflow runtime behaviors to support dynamic agent dispatch and correct control-flow graph wiring.

Changes:

• Adds a missing workflow.yaml implementing the Magentic orchestration pattern with dynamic agent selection.
• Fixes declarative runtime behavior for (a) expression-based agent name resolution, (b) initializing workflow state when JoinExecutor is the entry node, and (c) preventing successor wiring for terminator branch exits.
• Adds unit tests covering expression-based agent name resolution and terminator branch exit handling.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
python/samples/03-workflows/declarative/deep_research/workflow.yaml	Introduces the declarative Deep Research workflow definition with manager-driven dynamic agent dispatch.
python/samples/03-workflows/declarative/deep_research/main.py	Simplifies workflow YAML path resolution to always use the local workflow.yaml.
python/packages/declarative/tests/test_graph_coverage.py	Adds tests for dynamic agent-name expression evaluation and terminator branch exit handling.
python/packages/declarative/agent_framework_declarative/_workflows/_executors_control_flow.py	Ensures workflow state initialization when JoinExecutor is used as the entry node.
python/packages/declarative/agent_framework_declarative/_workflows/_executors_agents.py	Enables expression evaluation for agent names in InvokeAzureAgentExecutor._get_agent_name().
python/packages/declarative/agent_framework_declarative/_workflows/_declarative_builder.py	Skips terminator actions as branch exits to avoid incorrect dual-edge wiring.

[TaoChenOSU]

I think the yaml workflow is supposed to be shared between Python and .Net. But let's wait for @moonbox3 's confirmation.

[moonbox3]

Yes, we should be using the DeepResearch.yaml sample (the same one that Dotnet uses from MAF's root) as configured previously.

[moonbox3]

This test looks like it's using the incorrect executor type. The true path is: EndWorkflow creates an EndWorkflowExecutor. This test seems to be working right now because both store _action_def, but we aren't validating the actual "prod path." How about:

from agent_framework_declarative._workflows._executors_control_flow import EndWorkflowExecutor

end_executor = EndWorkflowExecutor({"kind": "EndWorkflow", "id": "end"}, id="end")
end_executor._chain_executors = [end_executor]
exit_exec = graph_builder._get_branch_exit(end_executor)
assert exit_exec is None

[moonbox3]

This works, but could be improved, I think, if we add something like an is_terminator property on the base class. Just cause I see us reaching to the "private" _action_def attribute.

[moonbox3]

Yes, we should be using the DeepResearch.yaml sample (the same one that Dotnet uses from MAF's root) as configured previously.

[moonbox3]

Closing in favor of #4159.