feat(factory): First Pass on Quickstart Factory Skills for Green Field

[johnson2500]

Description

This PR is to add functionality to the Quickstart Factory to create a new Quickstart from a user's idea all the way to the final product.
This include most of the SDLC for creating quickstart projects.

Changes

Stage	Skill	Output	Description
1	rh-qs-discovery	data/prds/<slug>.md	Discovery of the user's idea and creation of a PRD
2	rh-qs-architect	data/designs/<slug>.md	Architecture of the quickstart and creation of a design document
3	rh-qs-scaffold	GitHub repo + CI/CD	Scaffolding of the quickstart and creation of a GitHub repo
4	rh-qs-implement	Working application code	Implementation of the quickstart and creation of a working application code
5	rh-qs-deploy	Helm chart + compose.yml	Deployment of the quickstart and creation of a Helm chart and compose.yml
6	rh-qs-document	README.md + docs/	Documentation of the quickstart and creation of a README.md and docs/
7	rh-qs-ship	PR URL + blog draft	Shipping of the quickstart and creation of a PR and blog draft

Considerations

This was tested on my machine using an already pre-built quickstart project to see how well it would do.
Please test this out on your own machines to see how well it works for you. This is NOT production ready and should be used accordingly.

Please leave comments and feedback on the PR to help improve the quality of the quickstart factory.

Known Issues

• Not the best test coverage out the box.
• We may have stale data when it comes to Openshift AI new features.
• We may be out of data on some docs especcially when it comes to the ai-architecture-charts repo this can change.
• We need to ensure after each step we are running tests and enuring no breakage occure.

Example outputs

The entire repo was created by the skills here:

• https://github.com/johnson2500/research-paper-assistant -> See final PR.
  Then entire repo was created by the skills:
• https://github.com/johnson2500/phd-research-assistant

[Hadar301]

I think that the default state should be:
1. Open a new rh-ai-quickstart/ repo - with an empty readme & license
2. Fork it under your user.
3. Work on the fork.

[Hadar301]

While running this skill, the a yml file was generated with the following lines:

pgvector:
    image: pgvector/pgvector:pg16
    ports:
      - "5432:5432"
    environment:
      POSTGRES_USER: postgres
      POSTGRES_PASSWORD: postgres
      POSTGRES_DB: semantic_router
    volumes:
      - pgvector-data:/var/lib/postgresql/data

For POSTGRES_PASSWORD:

1. The user should be able to set the value or be aware that we are using generic passwords.
2. Either way, it's better if we include the passwords under the .env file and create secrets later when deploying

[Hadar301]

Two comments about this stage:

1. The repo was cloned under /tmp directory by default, which probably happened during the scaffolding.
2. Since most of the code of the app is being written at this stage, I'd add a recommendation for the user to open their IDE.

[johnson2500]

I wasn't sure where to put the source code. We could also have this in the data/ directory like the other artifacts.

[Hadar301]

rh-qs-scaffold created pyproject.toml files. When running the implementation skill, it does not use uv for package management and Python virtual environment (.venv)

cd /tmp/vllm-semantic-router/packages/api && pip install -e ".[test]" -q 2>&1 | tail -5)
  ⎿  (eval):1: command not found: pip

cd /tmp/vllm-semantic-router/packages/api && pip3 install -e ".[test]" -q 2>&1 | tail -5

cd /tmp/vllm-semantic-router/packages/api && python3 -m venv .venv && .venv/bin/pip install -e ".[test]" -q 2>&1 | tail -3

[johnson2500]

Created issue for this.

[Hadar301]

I'd consider adding a skill (maybe with a sub-agent so the main context will stay clear), such that:

1. Building and pushing all the Containers that are defined in the Containerfiles
2. Check if the user is connected to a cluster and suggest deploying all the components e2e.
3. Report back to the main agent all the lessons learned so that it can document them and edit the code.

[YosiElias]

2. Check if the user is connected to a cluster and suggest deploying all the components e2e.

As part of the blueprint to RHOAI skill conversion, we also need an agent to deploy the project and debug deployment failures. I totally agree that this loop will save a lot of engineering time. However, we need to consider security, and we are checking the best options for how to restrict the agent while giving it enough freedom to deploy and debug on the cluster.
@johnson2500 @Hadar301 If you have any take on this, I'd love to hear it. Of course, once we find a solution, we'll share it so we can use the same agent.

[johnson2500]

2. Check if the user is connected to a cluster and suggest deploying all the components e2e.

As part of the blueprint to RHOAI skill conversion, we also need an agent to deploy the project and debug deployment failures. I totally agree that this loop will save a lot of engineering time. However, we need to consider security, and we are checking the best options for how to restrict the agent while giving it enough freedom to deploy and debug on the cluster. @johnson2500 @Hadar301 If you have any take on this, I'd love to hear it. Of course, once we find a solution, we'll share it so we can use the same agent.

Agreed. We may be missing an entire skill. We may need to make this an explcit step or ensure each stage has security considerations. I do not have ideas yet but we can brainstorm.

[johnson2500]

This is broken and needs a fix.