Suggestion: Make it more like a product.

[carter003]

Haha, I’m the one who started the first discussion🙂

Jokes aside, the following content may differ from your original design intentions.
So please take this simply as a suggestion from a stranger.

In my opinion, this should be a simple and easy-to-use product. Yes , a PRODUCT!

Principle 1: Point-to-Point

The interaction should be direct and focused.

That’s why I suggest designing the product around the concept of “skills.”
Each skill represents a clear, well-defined task, and the entire workflow is guided by AI.

For example:

  - Let the AI execute commands such as bash drift scan
  - Write dedicated sub-agents that analyze the scan results produced by drift

The AI orchestrates the whole process instead of the user manually chaining steps

In short, the AI should be responsible for execution, analysis, and guidance.

Principle 2: Less Is More

Markdown should be the single source of truth.

MD files can serve:

As output for humans

As structured input for AI

MD is everything.

We should minimize — or even completely avoid — additional tools.
The system should be AI-driven end to end.

Our ultimate goal is to make the project’s:

Paradigm

Structure

Workflow

Correct, consistent, and unified.

To achieve that, we should:

Use the fewest tools

Use the fewest commands

Do only what is necessary to reach the goal

This is also why I believe a dashboard should not be our first choice.

Further Thoughts: Directory & Responsibility Design

I think we need to carefully design a clear and standardized directory structure.

What Is for Humans?

What Is for AI?

What data formats are needed, how to pre-filter the data through scripts, and how to provide the AI with the minimum number of tokens. Maybe to record, plan, or track TODOs and pending work for drift-self. This is similar to the init–loop command in oh-my-opencode, where an agent.md file is added to each important directory.