Well, starting from the bottom, let’s consider that Genie is not just an HA extension, but is an ecosystem by itself, and you could create automation in it (at the moment more complex going with the thingtalk language, less by text/voice), so the missing support of the HA one is also because we are trying to address more different features as possible, than have duplicates.
Consider also that integration with other ecosystems (like HA) could have missing parts/functions because: 1) is not our core priority, 2) genie is an open source project and so it would like to rely also on the community support, 3) the core team is a small group inside stanford lab, so is difficult to “add” and “extend” features at the same time;
Said so, going with my suggestion is not exactly a workaround because in a way or another, starting an automation is an on/off event, then we can talk about that the automations are available and exposed by HA, but if HA gave different possibility to achieve same result why don’t use it?
Then back on the begin of your post, is not Genie that expose something, but HA that expose endopoints as websockets where Genie attach with a specific skill
https://github.com/stanford-oval/thingpedia-common-devices/tree/master/main/io.home-assistant
So it will be just a matter of extending the actual base skill to interact with that service and then ad a specific skill to manage that kind of commands/functions.
Then the only problem I can see there is on creating the right dataset for the nlp training in a way that it couldn’t be misunderstood with other commands because it would be something like “start this automation” or “turn this automation on” (where “automation” could help) otherwise “say” as you wrote, is already referred to a capability (mentioned by Giovanni)
https://github.com/stanford-oval/genie-toolkit/blob/master/data/builtins/org.thingpedia.builtin.thingengine.builtin/manifest.tt.in
“announce” is better because not in the “say” dataset, and could be a specific skill as for example the “drop in” by alexa,