In the explicitly cued task-switching paradigm, two cues per task allow separation of costs associated with switching cues from costs of switching tasks. Whilst task-switch costs have become controversial, cue-switch costs are robust. The processes that contribute to cue-switch costs are under-specified in the literature: they could reflect perceptual priming of cue properties, or priming of control processes that form relevant working memory (WM) representations of task demands. Across two experiments we manipulated cue-transparency in an attention-switching design to test the contrasting hypotheses of cue-switch costs, and show that such costs emerge from control processes of establishing relevant WM representations, rather than perceptual priming of the cue itself. When the cues were maximally transparent, cue-switch costs were eradicated. We discuss the results in terms of recent theories of cue encoding, and provide a formal definition of cue-transparency in switching designs and its relation to WM representations that guide task performance.