v0.8.61 ui freezed after auto spawn seveval agent

[bruce6135]

Description

Steps to reproduce

1. begin plan mode, typing some words, it shows the basic plan, then I input more words to improve plan.
2. codeWhale is in plan mode, and can not change mode, it automatic think the detail of implement, I press Esc to stop is, and change to yolo mode.
3. It auto spawn two agent to do the job, few minutes later, the ui is freezed, can not operate any more

Expected behavior

Actual behavior

Impact

Environment

• OS: win 10 x86_64
• codewhale version: v0.8.61
• Install method:
• codewhale doctor summary:
• Version Information:
  codewhale-tui: 0.8.61
  rust: rustc 1.94.0 (4a4ef493e 2026-03-02)
• Model/provider: deepseek-v4-flash
• Terminal app: codeWhale
• Shell: windows's 终端

Logs, screenshots, or recordings

[github-actions]

Thanks @bruce6135 for the report.

This issue is staying open for maintainer triage. CodeWhale gets better because people bring us real edge cases from real machines, providers, regions, and workflows.

If you can add a reproduction, logs, version output, screenshots, or the provider/model involved, that makes it much easier for us to verify and harvest the fix. Maintainers may comment /lgtmi to mark recurring issue reporters as approved so this intake note is skipped next time.

[idling11]

@bruce6135 Based on your description, I believe there are three potential causes. I think the following steps could be taken to fix this:

1. AcceptAgent/AcceptYolo should not automatically send a "Proceed" message—user confirmation should be required before sending it.
2. Cascaded spawn protection: In addition to depth limits, the total number of sub-agents under the same parent should also be limited.
3. Sub-agent completion events should be rate-limited to prevent unbounded channels from overloading the context.

I plan to start addressing this after my other PR is merged. #3283

[Hmbown]

Retargeting into v0.8.63 from the release triage pass. This belongs in the same lane as the config/sub-agent/runtime-control work because it affects user trust, release stability, or first-run/runtime behavior.

[Hmbown]

This freeze on auto-spawning several Agents may be a symptom of the admission behavior tracked in #3318: beyond max_concurrent the spawn path hard-errors (crates/tui/src/tools/subagent/mod.rs:2300) instead of queueing, while the launch gate paces starts via a held semaphore permit (#3095, mod.rs:3366-3383). Worth checking whether the hang is in the rejection/permit-acquire path under a burst (e.g. many spawns racing the cap, or queued permits never draining). Cross-linking so the fix and the queue-and-drain redesign stay coordinated.