Suggestion: alternative command timeout approach

Claude can now choose a timeout for command executions. This can easily go wrong when it's unclear how long a command should take, e.g. if it's running a test suite of some kind.

An alternative approach that might work better is to match how humans do it - every N minutes, send the output of the command so far to Claude and ask it if it wants to wait longer or press Ctrl-C. This would allow Claude to react to progress information printed by a command, allowing a form of adaptive timeouts.