/build-repo

Your task is to build the repository according to the starter specification pattern.

Unless otherwise stated, all paths are relative to the base of the repository (the current working directory).

Repository Starter

The user will provide a file called prompt.md placed at the root of the repository.

This file will be a lengthy outline of everything that the user envisions for a project they wish to develop with your assistance.

Your task is to break up the information contained in this file by writing parts of it into certain parts of the repository.

Follow this order precisely.

1. Create A Starter Spec

Create an initial development spec at: spec/starter.md

The starter spec should translate the user's vision into a well-defined specification. The litmus test this spec must pass: is it sufficiently clear and well-directed that it would provide enough direction to an AI agent to begin developing the project?

You will be routing context data elsewhere, so consider excluding material from this file that can be captured in the context structure.

2. Create A Context File

In addition to providing instructions for how the project should function, the prompt the user provided likely contains what may be called "context data."

This context data might provide helpful background that you should use to ensure guidance is as useful as possible.

Save this content at context/index.md, rewritten in a precise, bulleted style:

- The user wishes to use Python for this project
- The user has identified React as a potential framework for the frontend but is open to alternatives

Do not repeat information already captured in the spec.

If the context information is large and would benefit from modularisation, create additional files within the context/ folder with descriptive filenames (e.g., context/technology-stack.md, context/ui-preferences.md).

3. Setup Wrap-Up

In CLAUDE.md you will find a placeholder:

{Project details}

Overwrite this placeholder with a summary of the project details to guide yourself during development.

4. Archive prompt.md

Move prompt.md into the archive/ folder to avoid confusion.

Once that is done, present the task completion and inform the user that the repository has been set up according to the model.

Speech-to-Text Awareness

The prompt.md file is often captured via speech-to-text input. When processing it, be aware of common transcription artefacts:

• Homophones: their/there/they're, to/too/two, your/you're, its/it's
• Technical terms: May be misheard (e.g., "Postgres" → "post grace", "OAuth" → "oh auth")
• Missing punctuation: Can affect meaning and sentence boundaries
• Ambiguous phrasing: Context may be needed to disambiguate intent

When you encounter unclear, contradictory, or suspiciously worded content:

1. Infer the likely intended meaning when obvious
2. Ask for clarification when ambiguous or critical to implementation
3. Do not assume the user made a logical error — consider transcription as the source