What
When Claude checks on an operation that is still running, such as a long shell command, the result now includes the last part of its output (output_tail). Before, it reported only that the operation was still running and its state.
- While the state is running, the tail is shown to the model as output.
- The loop that checks on the operation now also reads
output_tailoroutputfrom progress updates. - Error messages for an unresponsive or unbound host now add a note about the output.
- For in-progress shell command results, the internal record kept about each check now marks whether output was included.
Why
Claude can see what a long-running operation has printed so far while it waits. It can report progress or spot a problem before the operation finishes.
Something disagreesSomething we can check disagrees with this entry, or the writer said they could not settle it.
The writer flagged doubt
It is not clear which tool this applies to, or whether it is switched on for everyone.