-
Notifications
You must be signed in to change notification settings - Fork 3.2k
Description
@jayesh-tanna opened a sample PR (#44956) to demonstrate how services -- ML, in this case -- might want to define Copilot prompts. Since we have a monorepo with many services, it's important that these prompts are organized in a scalable, usable way that preserves functionality while avoiding cluttered usage, e.g. when specifying a prompt in a GitHub Copilot chat window.
When reviewing the images in the GitHub Copilot chat window, you’ll notice that pressing / displays all prompts. Given that the azure-sdk-for-python repository contains hundreds of libraries, this could result in a large number of prompts appearing when the user presses /. Similarly, the agents created by each library will appear in the agent popup in the same order. We can store instructions and skills within the appropriate subdirectory (e.g., .github/instructions/azure-ai-ml/someinstruction.md), while prompts and agents can be placed in the root directory (e.g., .github/agents/some_agent.md).
Proposal for this issue:
- We can distinguish prompts by name, as demonstrated in this pull request.
- Consider having common prompts, skills, or agents across libraries, as there are already some prompts and instructions in the .github folder.
