Since ~2 days ago, the chat panel for subagents launched via the Task tool never loads. Clicking on any subagent card opens a tab that shows “Loading Chat” indefinitely — both while the subagent is running and after it has completed. This affects all subagents, every time (reproduced consistently with multiple parallel explore-type subagents).
The subagents themselves execute correctly and return their results to the parent agent; only the viewer is broken. Before this started, I could open a subagent while it was running and watch its conversation stream live.
Possibly related: around the same date, some subagents started hanging indefinitely (one ran >1 hour producing no output, no tool calls, no edits, and had to be killed manually). Without the live view there is no way to detect these hangs early.
Steps to Reproduce
Ask the main agent to launch one or more subagents (Task tool, e.g. explore agents).
Click on the subagent card/tab while it runs (or after it finishes).
The tab shows “Loading Chat” forever; the conversation never renders.
Hey @Marcos_Campos
This is a subagent-viewer issue we’ve seen resurface recently, and I’ve reported your case to our team so it’s tracked with the current wave. There’s work underway on the subagent viewer, but it isn’t in a stable release yet, so updating won’t clear it right now (no need to chase newer versions).
A workaround that should help in the meantime: the tab gets stuck when you open a subagent while it’s still running. If you wait until the subagent finishes and then click its card (or click it again if you already have a stuck “Loading Chat” tab open), the full transcript should load. Subagents opened via resume also render normally.
On the separate issue you flagged (subagents occasionally hanging with no output for a long time), that’s a known streaming/buffering problem that tends to show up with Composer-family models and many simultaneous subagents. Two things that help until it’s fixed:
Pin a specific non-Composer model (e.g. a Claude or GPT model) on your subagents via the model field. See the subagents docs.
Run fewer subagents in parallel at a time.
cc @Duong_Do - I can see you’re hitting this too. I’ll follow up here as the viewer fix moves along.
Hi @mohitjain still facing the same issue and is not only with composer is with all the agents. this a stopper in some cases becase some agents get stuck an is no way to see if ther active o stucked. pls you help tu push this
Best way to tell if a subagent is actually working: watch its card in the parent chat, not the standalone tab. The parent card keeps updating live status (current step / todos) even while the child tab is stuck on “Loading Chat.” Running fewer subagents in parallel also makes the hang much less likely, and if a card shows no movement for a while, treat it as hung, stop it, and re-run.
Both the stuck viewer and the hang are issues we’re tracking. If you can grab a request ID from a subagent that hangs (three-dot menu → Copy Request ID) and drop it here, I’ll get it in front of the team. I’ll post back when there’s an update.
As observed, there is currently no way to determine the real-time status of the sub-agent, as it indicates “stopped” while the main agent shows waiting. This has occurred multiple times. Additionally, I have noticed that even when it displays as stopped, the sub-agent is still operational, which can be confusing.
