Everything visible as it happens

The transcript you see while a turn runs is not a curated highlight reel assembled afterward, it is the actual sequence of steps as they occur. Each tool call gets a one-line header the moment it starts, naming the tool and the one detail that distinguishes this call from any other call to the same tool.

Where a result has something immediately worth surfacing, a shell command's exit code being the clearest example, that appears directly beneath the header rather than being held back until the final summary.

Once the turn finishes, a short closing line recaps everything that happened in aggregate, files read, files edited, roughly how much changed, so you are never required to scroll back through the full transcript just to know what occurred. The exact format of both is covered in Outputs.

This live visibility is not a debugging feature bolted on for troubleshooting; it is the normal, default way every turn presents itself, whether the request is trivial or genuinely complex. There is no quieter mode that hides the step-by-step transcript in exchange for a single final answer.

In practice this means you rarely have to wonder what tCode is currently doing partway through a long-running turn, the shell command it is waiting on, the file it just started reading, is visible the moment it starts rather than only once the whole turn concludes.