[Bug] Resume can populate BaseAgentState for SequentialAgent and fail with ValidationError #5102

Required Information

Describe the Bug:
When resuming a resumable workflow that contains a built-in SequentialAgent, ADK can restore a placeholder BaseAgentState() for the workflow agent instead of a SequentialAgentState. The subsequent resume path then fails with a pydantic ValidationError when SequentialAgent tries to load its state.

Steps to Reproduce:

1. Install google-adk==1.28.0
2. Run the following minimal reproduction snippet:

from types import SimpleNamespace

from google.adk.agents import SequentialAgent
from google.adk.agents.invocation_context import InvocationContext
from google.adk.agents.sequential_agent import SequentialAgentState

workflow_agent = SequentialAgent(name="workflow_agent", sub_agents=[])
root_agent = SequentialAgent(name="root_agent", sub_agents=[workflow_agent])

ctx = SimpleNamespace(
    is_resumable=True,
    agent=root_agent,
    agent_states={},
    end_of_agents={},
    _get_events=lambda **_: [
        SimpleNamespace(
            author="workflow_agent",
            content=None,
            actions=SimpleNamespace(end_of_agent=True, agent_state=None),
        ),
        SimpleNamespace(
            author="workflow_agent",
            content=object(),
            actions=SimpleNamespace(end_of_agent=False, agent_state=None),
        ),
    ],
)

InvocationContext.populate_invocation_agent_states(ctx)
print("agent_states after populate:", ctx.agent_states)
print("end_of_agents after populate:", ctx.end_of_agents)

workflow_agent._load_agent_state(ctx, SequentialAgentState)

3. Observe that InvocationContext.populate_invocation_agent_states() stores BaseAgentState() for workflow_agent
4. Observe that _load_agent_state(..., SequentialAgentState) fails with a validation error

Expected Behavior:
Resuming a workflow with a built-in SequentialAgent should not fail with a state type mismatch. ADK should either:

• avoid synthesizing placeholder BaseAgentState for workflow agents, or
• synthesize the correct workflow-specific state type, or
• safely ignore placeholder state for workflow agents during resume.

Observed Behavior:
SequentialAgent resume fails with:

pydantic_core._pydantic_core.ValidationError: 1 validation error for SequentialAgentState
  Input should be a valid dictionary or instance of SequentialAgentState [type=model_type, input_value=BaseAgentState(), input_type=BaseAgentState]

Environment Details:

• ADK Library Version (pip show google-adk): 1.28.0
• Desktop OS: macOS
• Python Version (python -V): Python 3.13.7

Model Information:

• Are you using LiteLLM: No
• Which model is being used: N/A, this failure occurs before any model call

Optional Information

Regression:
Yes. I also reproduced the same behavior on google-adk==1.27.3.

Logs:

agent_states after populate: {'workflow_agent': BaseAgentState()}
end_of_agents after populate: {'workflow_agent': False}

Traceback (most recent call last):
  File "<stdin>", line 34, in <module>
  File ".../site-packages/google/adk/agents/base_agent.py", line 185, in _load_agent_state
    return state_type.model_validate(ctx.agent_states.get(self.name))
  File ".../site-packages/pydantic/main.py", line 716, in model_validate
    return cls.__pydantic_validator__.validate_python(
pydantic_core._pydantic_core.ValidationError: 1 validation error for SequentialAgentState
  Input should be a valid dictionary or instance of SequentialAgentState [type=model_type, input_value=BaseAgentState(), input_type=BaseAgentState]

Screenshots / Video:
N/A

Additional Context:
According to the resume docs, SequentialAgent should resume by reading its saved current_sub_agent state. However, in the installed package, InvocationContext.populate_invocation_agent_states() can synthesize BaseAgentState() for an agent that has content in event history but no explicit agent_state. If that agent is a built-in workflow agent such as SequentialAgent, the later resume path fails because _load_agent_state(..., SequentialAgentState) expects a workflow-specific state object.

This looks like an internal type mismatch between the synthesized placeholder state and the workflow-specific state expected during resume.

Minimal Reproduction Code:
See the snippet above.

How often has this issue occurred?:

• Always (100%)