runAgentCommand method

Future<ShellRunResult> runAgentCommand(String line)

Runs line in this (the user's current) interactive session and returns its captured output and exit code — used by the AI agent so its commands run in the real terminal shell (with a PTY, shared cwd/env and sudo credentials), streaming live and letting the user answer prompts.

Rejects if the session has ended or a command is already in flight. The command's output still streams to onOutput/onStderr (the user sees it) while it is also captured for the caller.

Implementation

Future<ShellRunResult> runAgentCommand(String line) {
  if (_ended) {
    return Future.error(StateError('session has ended'));
  }
  if (_inFlight || _agentRun != null) {
    return Future.error(StateError('a command is already running'));
  }
  if (line.trim().isEmpty) {
    return Future.value(const ShellRunResult(<int>[], 0));
  }
  final completer = Completer<ShellRunResult>();
  _agentRun = completer;
  _agentCapture = BytesBuilder(copy: false);
  _inFlight = true;
  onPassthrough(true);
  final command = _dialect.wrapCommand(
    line,
    interactive: _interactive,
    tail: _dialect.agentMarker(_marker),
  );
  _send('$command\n');
  return completer.future;
}