Starting a Project
Start with your business problem.
Voice record (and transcribe using Google AI Studio, TurboScribe, etc.) or write your user(s) flow, and create a Markdown file.
Create a project, add a specs
folder and PROJECT.md
in it.
mkdir specs && touch specs/THOUGHTS.md
Save to specs/THOUGHTS.txt
.
Prompt any coding agent:
Take transcription and write `specs/PROJECT.md`, focus on MVP only.
Prompt:
Read `specs/PROJECT.md`, add `Questions for Clarification`, with questions and default answers focusing on business, UI/UX, user journey, and similar details for MVP. I will edit answers if needed.
Check specs/PROJECT.md
, see the section Questions for Clarification
, and edit answers as needed.
Prompt:
We want reliable software, typed languages for backend/frontend apps only. API payload types should be generated from the backend. Deployment automation using infrastructure as code and GitHub-based CI/CD. We want end-to-end and integration tests, code format, lint checks through CI/CD. Read `specs/PROJECT.md` and create additional documents in the `specs/` folder where we need to capture business details like `Choices for XYZ field in API/UI` or technical details like `Deployment Plan`, `Backup Plan`, `Testing Plan`. Link the created documents to `specs/PROJECT.md`.
Prompt:
Read documents in `specs/`, create `CLAUDE.md` for Claude Code, link to documents in `specs/` as needed. Create `README.md` for Product Owners and general users.
I will paste Development Workflow
(see below)
into CLAUDE.md
; change it as you need.
Prompt:
Let's create a GitHub issue to build the backend and web app with a basic 'Hello World' API between them so we can manually test the apps and integration. Follow `specs/PROJECT.md` and linked documents for technical inputs. Mention in the issue: Types in API must be generated from the backend. Use the `gh` command for GitHub.
Work on the first issue using the prompts in Initiate Development