Clarification on bypass_state_injection usage in _process_agent_instruction #3790

I’m trying to understand the rationale behind the bypass_state_injection flag in the _process_agent_instruction function:

async def _process_agent_instruction(
    self, agent, invocation_context: InvocationContext
) -> str:
    """Process agent instruction with state injection."""
    raw_si, bypass_state_injection = await agent.canonical_instruction(
        ReadonlyContext(invocation_context)
    )
    si = raw_si
    if not bypass_state_injection:
        si = await instructions_utils.inject_session_state(
            raw_si, ReadonlyContext(invocation_context)
        )
    return si

Currently, it seems that the necessity of bypassing state injection is determined by whether the “instruction” is a string or a callable. In my use case, I am using a third-party tool to manage instructions dynamically, which means the instruction cannot be represented as a simple string—but that doesn’t necessarily mean it shouldn’t have state injected.

Could you clarify the intended use of bypass_state_injection and whether there’s a recommended approach for handling instructions that are not strings but still require session state injection?

Thanks in advance!