CLI: GPT-5.6 Sol starts tons of tasks, and they never finish

[loodvn (Lood)]

Where does the bug appear (feature/product)?

Cursor CLI

Describe the Bug

When using GPT-5.6 Sol recently, I’ve seen an increase in task (subagent) creation with most tasks staying stale and not completing.
See the attached screenshots - I had about 40 tasks this morning after an overnight run, most of which appeared to be stalled subagents. This happened with the pre-Aug 4th CLI version and continued after the Aug 4th update (I think). This hasn’t happened in my Grok sessions though.

Steps to Reproduce

Not sure how to reproduce sorry - I tried to spawn some tasks in a new session and the subagents closed fine.

Expected Behavior

Subagents should start, do their work, and then finish - the total number of tasks should reflect active subagents.

Screenshots / Screen Recordings

Operating System

Linux

Version Information

╭────────────────────────────────────────────────────────────╮
│ │
│ About Cursor CLI │
│ │
│ CLI Version 2026.08.04-aaa8809 │
│ Model GPT-5.6 Sol 272K Medium │
│ OS linux (x64) │
│ Terminal tmux │
│ Shell zsh │
│ Last Request ID 5231aebe-2d39-41e1-8cba-2ce1205d3b4b │
│ User Email [email protected] │
│ │
╰────────────────────────────────────────────────────────────╯
Could not copy to clipboard

For AI issues: which model did you use?

GPT-5.6 Sol [272K Medium]

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

[deanrie]

Hey, thanks for the detailed report with screenshots and the version.

Good news: the subagents themselves most likely did run and finish their work. The issue is with the task list, not with them. Lines like Subagent call_Iui, call_GB7, etc. are leftover placeholder entries from Task calls that got interrupted or retried mid-stream. When that happens, the line can get stuck in the Tasks pager with a “running” status and it won’t go away. k in the pager doesn’t stop foreground tasks, and /clear doesn’t remove them either. The only way to clear them is to restart the CLI session.

Why it’s more noticeable on GPT-5.6 Sol than on Grok: Sol starts a lot more parallel Task subagents, especially during an overnight run, so there are more chances for some entries to get stuck. It’s not model-specific behavior, Sol just puts more load on that path.

This is a known issue and we’re tracking it. I can’t share an exact ETA yet, but I’ll reply in the thread when there’s an update. For now, as a workaround, restarting the CLI session will reset the stuck lines.

Let me know if tasks still show as running after a restart, or if you notice that the subagents’ real work isn’t actually happening. Then we can dig deeper.

[loodvn (Lood)]

No worries, happy to restart periodically to clear the queue