Per rhn-support-bpratt – I think would be great to include things like API specs for the Konflux and Tekton CRDs for example so when we try to debug a pipelinerun failure it can better reason about if it's a spec problem

I think we would need to do several things for this:
1. Add these document sets to s3
2. Sync them with the konflux assistant ... or perhaps a new assistant?
3. For this set of docs, leave code blocks enabled – because we will want the LLM to have context about what a correct CRD should look like
4. Increase the embedding context window so that we can include entire CRD definitions inside document chunks
5. Switch to using mistral-small on HCM AI service because it has a much longer context window (64k tokens)
6. Adjust the prompt for this assistant so that it knows its job is to spot configuration errors.