Mirror Mirror · Webinar · Back to Basics

Claude Code: Back to Basics

Go from an empty folder to a working platform — by talking to your codebase.

Brady Hugins · Mirror Mirror

What it is

An AI that builds in your repo, not just chats.

Claude Code runs in your terminal (and IDE/desktop) right inside your project. It reads your files, writes code, runs commands, and explains itself — you stay the operator. The skill isn't coding; it's directing: say what you want, review what it proposes, ship what's right.

Your first 15 minutes

Empty folder → oriented + building

1 · Install + launch — npm i -g @anthropic-ai/claude-code, then cd your-project and run claude. Sign in once.
2 · /init — it reads the project + writes a CLAUDE.md (the project's memory: stack, conventions, how to run things).
3 · Explore — ask "walk me through this codebase" / "what does X do" before changing anything. Build a shared map.
4 · /plan your first change — describe a feature; it researches + proposes a plan; you approve; it builds. (More next.)
5 · /status — check where you stand (model, usage, config). Stay oriented.

The pieces that matter

Six components · start with the first three

CLAUDE.md ★
Project memory — decisions, stack, conventions. The #1 lever.
Plan mode (/plan) ★
Think before edit — research → plan → approve → build.
/status ★
Where you stand: model, context usage, config health.
MCP servers
Connect external tools/data (Airtable, GitHub, Stripe…).
Subagents
Spin up parallel/specialized helpers for big jobs.
Hooks + permissions
Guardrails — what runs automatically, what asks first.

Master CLAUDE.md + /plan + /status first; the rest you grow into.

/status

Know where you stand before you spend.

/status shows your model, account, config health, and how much of the context window you've used. Like checking your fuel + map mid-trip: it keeps you oriented so you don't run a long session blind or burn context on the wrong thing.

/plan — the operator's move

Think first. Edit second.

/plan puts Claude in plan mode: it researches your codebase, writes a concrete plan, and waits for your approval before touching a file. You catch scope + direction on paper — cheap — instead of in a tangle of edits — expensive.

Example: /plan add email login → it maps your auth, proposes the steps + files, you tweak/approve → then it builds exactly that. No surprises.

Building a platform

One loop, repeated

/plan → build → verify → repeat

Plan a slice, build it, verify it runs, write the decision into CLAUDE.md, check /status, go again. Big jobs → hand to subagents. The platform grows one verified slice at a time — and the system remembers how it was built.

The mindset

You direct. It does the typing.

Back to basics is one habit: say what you want clearly, let /plan show you the approach, approve the right one, verify the result. Start with one small feature today — plan it, build it, ship it — and the rest is the same move, repeated.

Keep going

Build it with us

Bring your own project to the cohort and we'll plan + build a real slice together. Course · cohort · membership at mirrormirror.roseinthegrove.com/learn/

1 / 9