When invoke_expression is called and no ready console exists (or the active console switched), the tool starts/switches a console but returns early with:
Started new console PID#... Pipeline NOT executed - verify location and re-execute.
This forces the AI model to make a second identical call just to execute the command. In practice, the second call always sends the exact same parameters — the "verify location" step adds no value since the pipeline typically includes its own cd.
Expected behavior: invoke_expression should start/switch the console and then execute the pipeline in a single tool call, with console context info prepended to the result.
I have an implementation ready, could I submit another PR?