I'll ask some questions here about your experience with prompting. I don't know the answers, but perhaps we could do some experimenting. (Is it possible to turn on a thinking mode in CoPilot so we can see how it is responding to the prompt - a quick google implies that it might be, but I'm not sure.

Anyhow, my question is around "Avoid". I wondered whether "Only make changes to existing content if necessary for clarity or coherence when adding the new content." would be more forceful?

I believe this depends on which model you're using, and more recent models tend to be less sensitive to the positive-negative prompting distinction as long as the prompt is clear. Copilot and others generally default to the more recent models, but contributors can manually select their preferred model and we can't fully predict how people will use it. Intent may also matter—is the specific instruction intended to be more about what to do or what not to do? All that to say: I'm not sure, but let's keep an eye on this to see if it is behaving the way we want or if we should flip the phrasing here.

The reasoning models do report on their steps (with varying levels of specificity and accuracy), so we can definitely experiment here.

Is the idea that this is for internal use? Is there a way that we can encourage/force contributors to use this when adding new content?

Primarily for internal use I think, but I added a line to the Jira ticket about developing best practice guidelines for Copilot/other AI assistants in our repo, and when we have that we can use it to point contributors towards the slash commands we want them to use.

Will this also use the copilot-instructions file to help it do this?

Yes! All Copilot usage in this repo will refer to the copilot-instructions file automatically.

These are all zero-shot prompts. Should we look at adding some one-shot prompts (perhaps later)? Perhaps something to put on our backlog?

Yes, worth considering! My experience is it's often best to start with zero shot and then add in examples on an as-needed basis if we find that it isn't picking up the instruction as expected.

Safer to say use the structure in the current document. I'm not sure we will be creating a completely new set of release notes without guiding copilot more from the request prompt?

Hmm, true. Or something like "Use the structure in the current document. Raise any inconsistencies in the chat, and cross-reference them against templates/release-notes-template.md."

I think I would leave this out for now? There are too many options and different styles.
Perhaps brainstorm it with the team and try to come up with a good description of what a release note looks like. Perhaps have two separate sets of instructions, one for Studio Pro release notes, and one for all the others.

Good point—I'll hold off on this one. I'll also add a note to the Jira ticket to revisit all of our existing templates and make sure we're happy with the way they capture our docs and various use cases. It would also be good to more fully include our style guides from Confluence directly in the Community Tools part of the repo and see if we want to simplify, clarify, or update those at all too.