Using prepend strategy for constitution template

[aki59]

I have created a custom component testing based preset and in that I have written a constitution template that will use prepend strategy on the core template

provides:
     templates:
                - type: "template"
                - name: "constitution-template"
                - file: "templates/comptest-constitution.md"
                - strategy: "prepend"

My comptest-constitution.md looks something like this:

# component testing suite constitution

## I. Architectural Principles

### Article 1 : Behaviour-driven Development
** Gherkin style Syntax**: All tests use given-when-then patterns.

## II. Adapting this constitution for new projects

###Step 1: file structure setup
domains/<domain>/tests/
|-src/test/java

There are much more to it. But when I am running speckit.plab command it never take into account the folder structure and shows in plan some random folder name to be created.

I am using gpt-5-mini for this, so not sure if that can be the issue.

[mnriem]

Hi @aki59, thanks for the question!

The issue is a YAML formatting bug in your preset.yml. Each property is written as a separate list item (- key: value), but they need to be keys of the same dict — only the first property gets the -:

What you have (incorrect):

provides:
  templates:
    - type: "template"
    - name: "constitution-template"
    - file: "templates/comptest-constitution.md"
    - strategy: "prepend"

YAML reads this as four separate objects, each with a single key — so none of them have type, name, and file together, which is required.

What you need:

provides:
  templates:
    - type: "template"
      name: "constitution-template"
      file: "templates/comptest-constitution.md"
      strategy: "prepend"

Once that's fixed, your constitution content will prepend correctly to the core template.

Regarding the folder structure not appearing in the plan: the prepend strategy gives the AI agent your text as context, but how faithfully it reproduces directory layouts depends on the model. With gpt-5-mini you may see it take liberties — try a larger model if the output still isn't matching your template after the YAML fix.