Azure / Azure/azure-sdk-tools

Utilize Copilot Instructions, Agents files in SDK repos for AzureSDK MCP

Open
#13,679 0 comments 0 reactions 0 assignees View on GitHub
AzSDK Tools Agent azsdk-cli
Dominant language
C#
Stars
135
Forks
260
Avg merge
3d 1h
Merged PRs (30d)
143

Description

Many (all?) of our SDK repos already have a trove of Copilot instructions, prompts, and maybe even agent definition that provide language-specific directory organization, test/sample instructions, etc., that seem best left to the SDK repos. Not only are they easier to update than in a centralized location, but they facilitate GitHub Copilot for the web, in VSCode, and in other use cases outside of `azsdk`. The language SDK repos become the central place for all things Copilot for any use case.

Case in point: since Azure/azure-sdk-for-rust isn't fully onboarded yet (we need to be closer to GA, I'm told), I did a little refactoring of existing Copilot instructions and wrote up some new Copilot instructions for ARM crates (something we're only starting to tackle officially in the public repo). After a couple iterations of tweaks, I have [something working](https://github.com/Azure/azure-sdk-for-rust/pull/3577) pretty well, but it will be undergoing some refactoring to split into 2 or 3 different PRs to isolate concepts and future `git log`ing.

For context, @ronniegeraghty asked me to open this and attach my [Copilot chat log](https://github.com/user-attachments/files/24780304/chat-keyvault-mgmt.json) as a source of some ideas and discussion. I think there's a good synergy of what we can consolidate into the `azsdk` tool and how we can utilize language-repo content in that orchestration. /cc @maririos

Contributor guide

Open the contributing guide

Research direction

Start by reading the azsdk tool and the Copilot material in Azure/azure-sdk-for-rust, including PR 3577 and the attached chat log. Compare the language-specific instructions, prompts, and agent definitions across SDK repositories. Done should be defined as an agreed orchestration approach and a documented set of repository content the tool can use.

Written by the indexing model from the issue text.

Assessment

Domain
tooling
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.