Chat orchestration

[Manu404]

Select Topic Area

Question

Body

TLDR
How can I orchestrate copilot chat usage by code ?

The yapping-details
I'm using copilot for personal projects on which I work almost totally alone. (and totally alone if we put the non-technical persons aside)
The usage I have of it is like a "performance static code review".
I ask the tool to identify possible performance improvements, then filter what is relevant. I don't care about anything else than performance improvements coming from it.
About 10% to 20% of the advices are legit. By that I mean relevant to me and not already done in a way the tool doesn't understand.
And maybe 5% are the impactful ones when benchmarking in various context under various loads etc.
But I've managed to get some real meaningful performances improvements from those 5%, so not a lot, but impactful enough to consider the hassle worth it in the end when working totally alone on a project without any external technical perspectives available on it.
For now, for each "scope level" I have to ask the tool to process my code (at method level, class level, etc) as some suggestions appear when asked at some level only.
I would like to be able to automate all that, so in the end I would be able to "press a button" and trigger all the reviews I programmed, let the tool work and come back later to check the "report".
I would like to not have any reliance on GUI, nor human interactions.
I would like something close to output to console and pipe to a file in the worst case.
I would also like to be able to say "that thing, you already suggested it, don't do it anymore" with a possibility to override that when I want a "full analysis", but this might mostly be on my side to handle that, not at the copilot level, I don't mind it.
I don't care about the environment, I don't want this to happen in a specific IDE or CI/CD or OS or else, just happen somewhere and I'll adapt, as long as it's not costing me an arm or my next newborn child in sacrifice.
Another "sweet" use case, would be to go even a step further by automatically implementing each suggestions and using my automated benchmarks to have the meaningful improvements from each suggestions already "sorted out" to gain a bit of time in the analysis of the copilot output. But that could also be done on my part if I want to, once I can orchestrate chat usage. (I'm not even remotely considering having those changes automatically integrated or anything, don't worry, just some form of smoke testing to focus on the ones who already provided better measurements through that process first)

Due-diligence
Ofc I searched this, but the only solutions I've found are kinda "hackish" to say the least. If doing their solutions end up costing more time in maintenance and/or lower the level of confidence and/or consistency toward that process, to the point of not even being sure the right and/or whole code got analyzed, I'm not interested.
Now, I would also "understand" why that kind of API would not be made public, or public in the 10$ tier, as it can end up generating an incredible amount of load if we all start to do that compared to the 10$ we pay.

[Manu404]

Pivoted towards a locally running model, all the control I could wish for, slow af but not concerned about response time, run those during downtime, so works for my context and needs.