there is no need to

map

the property, since Gradle automatically gets the output from the task. also you can do all of this with just ad-hoc tasks, and even make it more general and track input properties without much work:

import groovy.json.StringEscapeUtils.escapeJava

val generateBuildConfig by tasks.registering {
    inputs.property("PROJECT_VERSION", provider { project.version.toString() })
    val outputDir = layout.buildDirectory.dir("generated/source/buildConfig")
    outputs.dir(outputDir).withPropertyName("outputDir")
    doLast {
        outputDir.get().file("my/package/BuildConfig.kt").asFile.apply { parentFile.mkdirs() }.printWriter().use { writer ->
            writer.println("package my.`package`")
            writer.println()
            writer.println("public object BuildConfig {")
            for ((key, value) in inputs.properties) {
                writer.println("""public const val $key: String = "${escapeJava(value.toString())}"""")
            }
            writer.println("}")
        }
    }
}

kotlin {
    sourceSets {
        commonMain {
            kotlin.srcDir(generateBuildConfig)
        }
    }
}

of course, making a custom task type would allow you to make it cacheable, but that may only be worthwhile if it's reused