Exiting plan mode not reliable with new GPT-5.6 models

### Describe the bug

When creating a plan with the new GPT-5.6 models (doesn't matter which one), often the interaction ends like this:

`Plan saved to plan.md, with implementation tasks tracked in the session database.`

And then nothing happens, what is expected is that the user is prompted with how to execute this plan, but this almost never happens. What works is to do prompt something like:

`Present me the plan, and ask me to implement with autopilot`

And after that implementation works as expected.

Other models (GPT-5.5, Anthropic, etc) do not have this issue.

### Affected version

`1.0.71`

### Steps to reproduce the behavior

See above

### Expected behavior

See above

### Additional context

N/A

No files, tests, or entry points are named. Reproduce the failure in version 1.0.71 with a GPT-5.6 model, compare it with GPT-5.5 or Anthropic, and trace the plan-mode exit; done means the expected implementation prompt appears reliably after saving the plan.

Written by the indexing model from the issue text.