Extended thinking after tool call results

[StructByLightning]

Currently extended thinking only happens once at the start of a turn. When tool calls return, the model continues generating directly without another thinking phase.

Claude Code handles this differently by giving the model a thinking block after each tool result comes back, so it can reason about what it found before deciding what to do next. This consumes more tokens, but makes a noticeable difference in quality for multi-step tasks where the model needs to adjust based on what tools return.

Would love to see Continue support this pattern: thinking → tool call → thinking → tool call → etc, rather than thinking → tool call → tool call → tool call.

[StructByLightning]

Oddly enough, it seems like sometimes the desired pattern does happen:

but then in the same agent flow it doesn't always happen:

I was unable to intentionally trigger multi thinking when prompting the model directly, and it seems to only happen sometimes. So, since it seems to already be supported, it would be nice if there was a setting to force it to always happen