Opencode crashing mid execution losing reference to the session

[mscipio]

Hello! I finally updated after a few days and jumped from 6.26 to latest and since then I have experienced a lot of mid execution hanging of opencode that results in the session suddenly crushing out.

I don't really have a log to share (I think) because the only thing I can see is a pop-up in the top right corner of my terminal windows (I am inside vscode) that says that opencode could no longer find the session it was working on.

I am not sure it is related to opencode swarm (even though this is literally the only plugin I use and have installed) but I thought of mentioning. Hopefully someone here has some input on if I am doing something wrong or if it is indeed a bug related to the swarm.

Please let me know if you think of anything I could share to help debugging the issue.

Thanks a lot for all the work on this project!

[zaxbysauce]

Hey, thanks for reporting this and for being patient. I've been digging into this one pretty deeply and I'm fairly confident the root cause is upstream in OpenCode's SSE event handling — when a client disconnects or there's backpressure on rapid bus events, the writeSSE call throws unhandled and kills the session. I have a fix PR open upstream (anomalyco/opencode#18538) for this.

On the swarm side, v6.33.1 just shipped with a few things that should help reduce how often you hit this: snapshot writes are now debounced instead of firing on every single tool call, and there's better diagnostic logging so if it does happen again we'll have a trail to follow. There's also a new /swarm reset-session command that clears session state without nuking your whole .swarm directory.

If you're still hitting this after updating, logs from the terminal output would be really helpful.

[zaxbysauce]

This has been addressed across several releases (v6.33.5 through v6.33.9). The root causes were a stale delegation reversion bug (|| instead of && in the identity check), rehydrated sessions getting immediately evicted, and a hook chain timeout that was causing concurrent state corruption. All three are fixed. The upstream SSE fix (anomalyco/opencode#18538) is still pending but the swarm-side amplifiers are resolved. If anyone is still hitting session loss on v6.34.0+, please open a new issue with logs.