Skip to content

Latest commit

 

History

History
53 lines (36 loc) · 3.55 KB

File metadata and controls

53 lines (36 loc) · 3.55 KB

Bootstrap — starting a new project

Five steps. Do these once, at the very start of every new app, before you build features. They drop the framework's rules into your project so your AI tools follow them automatically.

Not comfortable in a terminal? You don't need to be. Each step below has a terminal version and an "ask your AI" version. The "ask your AI" version is a prompt you paste to your coding tool and it does the typing for you. That's the whole spirit of this framework — you direct, the AI does the mechanics. Pick whichever you prefer; they do the same thing.


1. Create your project + repo

One app = one repo.

Ask your AI: "Create a new [type, e.g. Next.js] app called [name] and initialise a Git repository for it. Then connect it to GitHub. Walk me through anything you need me to click."

2. Pull in the framework files

Get AGENTS.md, CLAUDE.md, and env.example into your new project's root.

Terminal version (if you're comfortable):

npx degit bankkaroproduct/build-framework/templates ./_framework-templates
# then move AGENTS.md, CLAUDE.md, env.example out of _framework-templates into your root

Ask your AI version: "Pull these files from the GitHub repo bankkaroproduct/build-framework (templates/ folder) into the root of this project: AGENTS.md, CLAUDE.md, and env.example (save the last as .env.example). Don't change their contents."

Or simplest of all: open the build-framework repo on GitHub and copy-paste the files in by hand. Any of these is fine.

Also grab templates/checkpoint.md — you won't fill it yet, but you'll want it once you finish your first milestone (it's how you hand state to a fresh thread when this one gets slow — see references/08-context-and-thread-management.md).

3. Fill in AGENTS.md

Open AGENTS.md and fill the top section: project name, what it is, Real-data or Demo (if unsure → Real), hosting, database. Thirty seconds. Now every AI session knows your project and your rules.

4. Lock down secrets

  • Copy .env.example to .env.local and put your real values there (this file is never committed).
  • Make sure your .gitignore contains .env, .env.local, .env*.local.
  • Put real secret values in .env.local (never committed) and in your hosting provider's Environment Variables settings (Preview + Production separately).

Ask your AI: "Copy .env.example to .env.local for me to fill in. Make sure .gitignore ignores all .env files so I can never accidentally commit a secret. Then tell me, in plain English, which secret values I need to go get and where to paste them in my hosting dashboard."

5. Write one paragraph of memory

In the "Project memory" section of AGENTS.md, write 2–3 sentences: what this app is and who uses it. Update it as you go. This is what saves you from re-explaining your project every session.


You're set

Now follow the loop in FRAMEWORK.md: Classify → Plan → Build → Verify → Ship-behind-a-gate. Your tools already know the safety rails because they're in AGENTS.md.

When in doubt about any step, the detail lives in references/:

  • 01-project-setup.md — setup specifics
  • 02-orchestration.md — directing the AI, handoffs, reviewer loop
  • 03-verification-shipping.md — how to verify, the production gate
  • 04-safety-rails.md — the safety rails in full (incl. security baseline + safe-write checklist)
  • 05-incident-escalation.md — when things break, when to call an engineer
  • 06-existing-project-mode.md — applying the framework to a project that already exists