Rubber-duck launches 4 critics every time

### Describe the bug

When I ask an agent to rubber-duck or if a subagent decides to rubber-duck, I without fail always get 4 rubber-duck instances. They all launch at separate times but consistently show up eventually. The all arrive at same conclusions so they don't add any extra value. I asked copilot why it launched four and it said it only called the tool once and there is no way to specify how many instances launched. It thought that perhaps each one was too slow to start responding so the code tried to launch another as a backup and so on until there were four.

### Affected version

1.0.22

### Steps to reproduce the behavior

Ask an AI to rubber-duck something. Watch as four background tasks slowly accumulate.

### Expected behavior

Only one rubber duck instance should launch at a time/per-request.

### Additional context

- Windows Terminal on Windows

Start by reproducing the issue from the rubber-duck entry point described in the report on Windows Terminal. Trace how a single request launches background tasks and check whether slow responses trigger additional launches; done means each request produces only one rubber-duck instance.

索引モデルが issue の本文から書いたものです。