We present an approach to children's programming inspired by the semiotics of comics. The idea is to build computer programs in a direct and concrete way by using a class of signs that we call contextual signs. There are two aspects that distinguish contextual signs from other sign systems used for programming. The first is that the signs are displayed in the immediate visual context of the object that they refer to. The second is that the signs are used to illustrate actions and properties in a way that is directly perceivable by the user. We argue that these two properties make contextual signs a promising high-level approach for building systems that are rich in dynamic properties, such as the ones that children often like to build.