Hi team, I found a possible race condition in AgentActivity where _speech_q is modified without using the class lock (heappush/heappop in different tasks). Can you confirm if this should be protected with the internal lock or if there’s an expected pattern here? Happy to open a PR once confirmed.

I think it would be best to open an issue on GitHub in the repo in question, that would give more context and it would need someone deeply familiar with the code to comment. Especially if you're seeing issues with your app that are fixed by your update.

Thanks, Darryn. I’ll open a GitHub issue with details and context so it can be reviewed properly.