Good practice?

[dlukt]

Right now, using Antigravity, after I went through all the Epic creation etc, I'm doing

1. create-story
2. dev-story
3. code-review
4. goto 1.

But I do all of those in a new session.

How do you do it?
I'm thinking, since 1. is ran, might as well run 2. in the same session, since context already exists. However depending on the model a warmed up context leads to worse results.

What is your experience?

[bmadcode]

they are separate and meant to be separate context for the sole reason minimization of the context for the dev agent to be able to code. as your project grows, the dev agent is going to need more and more available context to implement code properly. If you already exhausted the context reading lots of documents with a lot of content not related to the story at hand, it will have less context available for the actual coding task. the whole purpose of create-story is to generate a single file the agent needs to load.

but it does not hurt for you to experiment and learn what works best in what scenarios.

[jmaurelli]

I have learned to implement one story at a time. The dev agent will write notes and feedback into the story upon completion. Then when the second story is implemented, the next dev agent with a new context window will reference the previous story notes and feedback.