fix(cartesia): run the wrapped agent exactly once per event

[abhay-codes07]

What

SupermemoryCartesiaAgent.process() could run the wrapped agent twice for one event.

The whole method body sat in a single try/except whose fallback re-invoked self.agent.process():

try:
    ...memory enrichment...
    async for output in self.agent.process(env, event):
        yield output
except Exception as e:
    logger.error(f"[Supermemory] Error in process: {e}")
    async for output in self.agent.process(env, event):   # runs the WHOLE agent again
        yield output

The fallback exists so a memory failure never blocks the agent — but because the agent's own iteration lived inside the same try, an agent error raised mid-stream (LLM connection drop after chunks were already yielded) triggered the fallback too. The caller received a second, full response concatenated onto the partial one, plus a duplicate billable LLM call. This applied to every event type, not just UserTurnEnded.

Fix

Scope the guard to the memory enrichment/injection/storage block only, then run the agent iteration exactly once after it:

• enrichment failures are logged and absorbed, and the agent still runs — same guarantee as before
• agent errors now propagate to the caller instead of triggering a duplicate run

No changes to the enrichment, prompt-injection, or storage logic themselves.

Testing

New tests/test_single_agent_run.py streams from a fake inner agent that dies mid-stream:

• each chunk is delivered exactly once, the inner agent's run counter stays at 1, and the error propagates — both mid-stream tests fail against the previous implementation (verified: run counter hits 2 on the old code)
• a third test pins the intended behaviour that an enrichment failure still lets the agent run exactly once

python -m pytest tests in the package — 4 passed.

cc @MaheshtheDev

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.