setting include_content to 'none' of a Agent with some tools within a loopAgent result in Pipeline break

[Opty0602]

So I am building a pipeline, which is like
ManagerAgent -> AgentTool ( LoopAgent( solution_agent with a tool to retrieve solution + Validator Agent ] )

I am setting include_contents = 'none' in both of the sub agents in LoopAgent, cause i want both of the Agents to be run independently.

i am hydrating my states with necessary information for my subagents to be capable enough to produce an output.

but when my loop is on its second iteration, the pipeline breaks after solution_agent gets the response from retrieval tool (as i am explicitly prompted my agent to always call the tool) and the error i am thrown is

google.genai.errors.ClientError: 400 INVALID_ARGUMENT. {'error': {'code': 400, 'message': 'Unable to submit request because it must have a text parameter. Add a text parameter and try again. Learn more: https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/gemini', 'status': 'INVALID_ARGUMENT'}}\

Please help !!

[AurYou]

same error here !

[Opty0602]

same error here !

Ahh finally, someone who can relate with me, do you have any minimal reproducible example, I tried to get one but no can do , idk why.

[michael-t-alexander]

I have the same error when using include_contents="none" but with a different message:

google.genai.errors.ClientError: 400 INVALID_ARGUMENT. {'error': {'code': 400, 'message': 'Please ensure that function call turn comes immediately after a user turn or after a function response turn.', 'status': 'INVALID_ARGUMENT'}}

I haven't managed to work out a minimal reproducer, but I suspect the problem originates from here, which selects which Events to include in the contents passed to the Gemini API when include_contents="none".

I have a workaround though. Before the agent that causes the problem, I insert in the sequence a dummy agent that just yields an Event with some text content. Create a custom agent like:

from typing import AsyncGenerator

from google.adk.agents import BaseAgent
from google.adk.agents.invocation_context import InvocationContext
from google.adk.events import Event
from google.genai import types

class MessageAgent(BaseAgent):
    message: str

    async def _run_async_impl(self, ctx: InvocationContext) -> AsyncGenerator[Event, None]:
        yield Event(
            author=self.name,
            content=types.Content(
                role="model", parts=[{"text": self.message}],
            ),
            branch=ctx.branch,
        )

Create an instance like msg_agent = MessageAgent(name="msg_agent", message="I will now do the next step") and insert it in sequence before your agent(s) that have include_contents="none". This seems to ensure that the contents that gets sent to the Gemini API is valid.

It works for me anyway, so worth a shot. Gonna see if I can work out a proper fix in any case.

[Jacksunwei]

Thanks for reporting this issue. I've investigated the root cause and wanted to share my findings:

Relationship to PR #2961:
PR #2961 fixes a very similar issue in _get_current_turn_contents() for ParallelAgent scenarios. The fix adds more robust event filtering via _should_include_event_in_context(), which may also resolve this LoopAgent issue since both stem from the same logic in contents.py:441-449.

Unable to reproduce:
I attempted to create a minimal reproducer based on the issue description but was unable to trigger the exact error with the current codebase (ADK 1.16.0, GenAI SDK 1.46.0). The test completes successfully without errors.

Request for help:
To investigate further, could you please:

1. Provide a minimal code sample that reproduces the error consistently

2. Capture the exact LLM request that triggers the error by running with verbose logging:

   adk web -v path/to/agents_dir

   This will show the outgoing GenAI SDK requests in the logs, which can help identify what invalid content is being sent.

3. Share your versions:

   • ADK version (pip show google-adk)
   • GenAI SDK version (pip show google-genai)

Once PR #2961 merges, it would be helpful if you could test again to see if it resolves your issue.

[Jacksunwei]

Closing this issue for now since we cannot reproduce it with the current codebase. If you can provide a minimal reproducer or additional details, please feel free to reopen or create a new issue. PR #2961 may also address this when it merges.