Common Traps — az prototype

[lukemurraynz]

I gave az prototype a go on the weekend, pretty nice! Great to quickly get something going!

Thought I would share some of the traps I ran into (I turned them into a Skill to prevent re-occurance):

https://gist.github.com/lukemurraynz/05e314141289af3f53db4b8413b9c692

Some of them were client-facing (i.e., my Terraform version was out of date), and my Copilot Pro Plus plan doesn't include GitHub models - so failed back to Azure OpenAI endpoints, but a big one for me was:

• az prototype did not seem to NOT read your repo's .github/instructions/ or .github/skills/ files

Hopefully, the common-traps file is of some use. Overall pretty nice - once I got over my model issue and needing to use Azure OpenAI, it fast-tracked the design - also kudos for the in-built CAF naming conventions!

[a11smiles]

@lukemurraynz thanks for the detailed write-up. I will address the issues with fixes in the next couple of days.

Question: I understand the TF "issue". Recommendations or thoughts on how to mitigate? What's a plausible solution for our customers?

[lukemurraynz]

@lukemurraynz thanks for the detailed write-up. I will address the issues with fixes in the next couple of days.

Question: I understand the TF "issue". Recommendations or thoughts on how to mitigate? What's a plausible solution for our customers?

Just thinking outloud, could you let AZD do it as a baseline? I know the AZD handles the Bicep/Terraform deployment for new projects, etc, during an AZD UP. I wonder if this could be offloaded back to AZD, instead of having to redo the logic in both places, or get the team to align on the minimum version, unless an exception can be made (ie in the AZD environment variables).

[a11smiles]

the problem is, i submit the Terraform version to the AI API so that the TF modules that are returned are valid for the version. But, I could fallback my own version to AZD, i guess. However, I wonder how many customers are using dev containers.

[lukemurraynz]

the problem is, i submit the Terraform version to the AI API so that the TF modules that are returned are valid for the version. But, I could fallback my own version to AZD, i guess. However, I wonder how many customers are using dev containers.

It's easy to fix, as a user (upgrade the Terraform version/feature as either part of the devcontainer configuration or post-implementation script, etc), as long as the azd prototype does an init and fails fast when the version isn't right - I wouldn't spend a lot of time on it personally. I think in my use case, I had an outdated devcontainer version with an older Terraform configuration.