Need help about dev life cycle

[lhn153]

Currently, I’m trying ai-devkit with Cursor/Antigravity and have a question. For example, when I run the command "/new-requirement Some feature", the devkit generates a markdown file. If there are items at the end of the requirement that need clarification, should I edit the markdown directly, or should I respond to the agent in chat to clarify them? Is it the same the other way around for degisn, plan, test, ...?

[codeaholicguy]

Hi @lhn153, thank you for your question.

If there are items at the end of the requirement that need clarification, should I edit the markdown directly, or should I respond to the agent in chat to clarify them?

In an agentic workflow, I recommend responding in chat.

Let the agent drive the clarification loop. It will keep asking questions until the requirement is clear enough to move forward.

When you run /review-requirement or /review-design, the agent automatically surfaces the remaining open questions.

Is it the same the other way around for degisn, plan, test, ...?

Yes. The same principle applies across all phases. Clarify in chat, let the agent update the artifact.

Tip: If you use the dev-lifecycle skill, it automatically triggers /review-requirement or /review-design after generating artifacts. In most cases, you only need to type continue and answer the questions in the chat window. Here is the doc.

Let me know if you have any further questions.

[longlhdev]

I think the your requirement skill should be enhanced, maybe apply some skills like brainstorming to clear the input before do anything

[codeaholicguy]

That can be a good idea.

I agree that sometimes jumping straight into structured requirements without expanding the problem space can limit the quality of the output. A lightweight brainstorming phase before formalizing requirements might help surface assumptions, edge cases, and alternative approaches.

Do you have any concrete example where the requirement or design output felt weak or incomplete?

For example:

• Was the scope misunderstood?
• Were edge cases missing?
• Did the design feel too narrow or too implementation-driven?

If you can share one or two specific cases, it would really help us identify whether:

• We need a dedicated brainstorming skill before requirement,
• Or we just need to enhance the clarification loop inside the current requirement skill.

I prefer improving based on real failure cases instead of optimizing in theory. That way we enhance the system in a targeted way, not just add another phase because it sounds right.