Quite probably, your invented examples Infans lavandus clamabat and Urbs nobis capienda militiam novam paraverat would sound quite odd to a native speaker of Latin.
Note that the attributive use of verbal adjectives in -nd- was very restricted in Latin. As pointed out by Pinkster (2015: 998), "the attributive use of gerundives is almost entirely restricted to those that are derived from verbs denoting a personal evaluation (especially, the 'verba afectuum'). So one finds contemnendus, laudandus, mirandus, but not aedificandus". Note that your invented examples do not fall under this category, unlike the ones from A&G reported by TKR. (Joonas, in case you have access to the following book: Aalto, P. (1949). Untersuchungen über das lateinische Gerundium und Gerundivum. Helsinki: Suomalainen Tiedeakatemia, a list of attributively used gerundives can be found on pages 102-103).
According to Vester (1991: 306-307), the adjectival nature of gerundives is basically reduced to the attributive use, which is quite limited (it is only a bit productive in classical poetry). Putting this restricted attributive use aside, Vester (1991: 307) concludes "the gerundive then hardly shows prototypically adjectival behaviour, probably because its semantic content denotes action or process."
Furthermore, it is also worth pointing out that, in striking contrast to participles (cf. patria expulsus fugit / puer flens abiit), the subject predicative use of gerundives was also (generally: see my relevant remark below) excluded. E.g., see Vester (1991: 302): "it is at least clear that the gerundive does not occur as a Praedicativum in the nominative case, i.e., in agreement with the Subject"). Cf. the well-formation of the so-called "Subject complement" use of gerundive (e.g., Epistula est scribenda).
According to Vester (1991: 302), "the only function in which there seems to be a real parallelism (sc. between participles and gerundives: Mitomino) is the dominant construction". However, even this parallelism, when examined in detail, is nicely shown by Vester to be illusory: for example, according to her, gerundives do not enter into (true) ablative absolute constructions. She makes the very important observation that an identity/coreferentiality between the first argument of the gerundive and the first argument of the main predication is required in gerundives but not in perfect participles. For relevant discussion, see this post.
Let me make a couple of relevant remarks. I think that Vester's generalization that "the gerundive does not occur as a Praedicativum in the nominative case" is essentially (and, indeed, intriguingly!) correct except for those cases that involve a passivization of the so-called 'object complement"/predicative use: e.g., Sicilia quae mihi defendenda tradita est (Cic. Ver. 5.188).
Finally, please note that your second invented example is doubly ill-formed: not only because of the first reason mentioned above but also because a "dative of agent" (nobis) is not possible in this non-verbal context (again in striking contrast with ablatives of agent with perfect participles: urbs a nobis capta militiam novam paraverat). For related discussion on datives of agent in non-verbal contexts, please see this post.
SOURCES:
Pinkster, H. (2015). The Oxford Latin Syntax. Vol. I. Oxford: Oxford University Press.
Vester, E. (1991). "Reflections on the gerund and gerundive". In Robert Coleman (ed.). New Studies in Latin Linguistics. 295-310. Amsterdam/Philadelphia: John Benjamins.
PS: I'm sorry if you find the grammatical terminology used above a bit confusing (e.g., cf. "attribute", "subject complement", "object complement", "praedicativum", etc). Pace Pinkster, Vester, and other functionalist linguists, I do also find it misleading. Fortunately, in my country (Catalonia) we do not use it.
