OpenAI Agents, Google ADK, and LangGraph HTTP templates are stateless — no conversation history across invocations

[aidandaly24]

Description

The OpenAI Agents, Google ADK, and LangGraph HTTP agent templates have no conversation history persistence. Every invocation starts with zero context — the agent cannot recall anything from prior turns within the same session.

For example, if a user says "my name is Aidan" and then in a follow-up invocation asks "what's my name?", the agent has no idea. Each invoke is treated as a completely independent interaction.

Steps to Reproduce

1. agentcore create --name test --defaults --framework OpenAIAgents (or GoogleADK, or LangChain_LangGraph)
2. agentcore dev
3. Invoke with {"prompt": "my name is Aidan"}
4. Invoke again with {"prompt": "what is my name?"}

Expected Behavior

The agent should recall "Aidan" from the previous turn within the same session, since both invocations share the same session context.

Actual Behavior

The agent responds with something like "I don't know your name" because it has no memory of the prior invocation.

Additional Context

This affects three of the four HTTP templates:

• OpenAI Agents: No session persistence mechanism — each invocation creates a fresh Runner.run() call with no prior context.
• Google ADK: InMemorySessionService creates a new session every invocation via create_session(), discarding prior turns.
• LangGraph: No checkpointer configured — the graph has no mechanism to persist state between invocations.

The Strands template is NOT affected by this issue — it reuses the Agent instance which accumulates messages internally (though it has a separate conversation bleed bug).