What is the best way to update / rebuild incrementally when new requirements are discovered

[fitzgeraldsteele]

I created an initial spec with /specify, completed the planning and tasks steps. Then oops...I realized I missed a requirement. What is the best way to incorporate changes into the spec? I know I can update spec.md directly, but it feels like it would be better to have the agent take my raw input for the changes I want to see, and blow them out to full blown specs like it did the first time through.

[raulduk3]

Each time you use the/specify command you branch off and the agent begins to develop a specification for a feature set.

It's up to you to integrate those features into the main branch.

Also, I believe new specifications are contextually informed about previous specifications (and by the overall constitution).

Mainly just play around with branching and merging working code in.

[soichisumi]

I also need to establish a method for modifying existing requirements.
Specifications often change over time, and in such cases, I believe it is more appropriate to modify the existing specification document (spec.md) rather than creating an additional one.

Should also related to this discussion: #52

[colinrees]

So the way I have been working (please anyone who knows better let me know.)

I put ALL my tech requirements in the constitution, I don't want to have to keep doing that for each requirement, I then pull out the features and do a /specify for each feature.. for example

/specify please create me bootstrap for the CRM system that I am developing.
In scope:

• A based UI with sidebar, top nav
• logging
• testing framework
• dev environment configuration

Out of scope

• Permissions
• Any functionality other than the above
• production deployment.

I did also need to put each of the prompts through the anthropic prompt optimiser (using Claude code here), to get them stop over generating; but I think the results are now pretty good.

Feedback on this approach appreciated.

Colin