What this rung really is
Level 2 is where you stop repeating yourself. At Level 1 you briefed the agent fresh every time — your voice, your do's and don'ts, the way you like things done, said again on every job. The move here is to write those standing rules down once, in a file the agent reads before every single task. From now on “follow my rules” is automatic, not something you re-explain.
The second half is just as simple: you make the agent show its plan before it touches anything. Instead of acting and hoping, it lays out what it intends to do, you glance at it, and only then does it go. Clear this and the agent stops being a clever improviser and becomes a reliable operator — same standard, same rules, every run.
Write the rules once — then stop repeating yourself
The move at this rung is standing rules: put your voice, your do's and don'ts, and where things get saved into one file the agent reads before every job — and make it show its plan before it acts.
This rung sits on everything you wired in rungs 1–1. You add written rules & a plan gate on top — you don't start over.
Do this rung, for real
Everything you need is here — no tabs to chase. First, the jargon this rung throws at you, in plain words. Then the steps, with the exact things to paste or say.
You write a CLAUDE.md once: “Always use British spelling. Never invent stats — leave a [VERIFY] tag instead. House voice is plain and direct, no hype. Save drafts in /drafts.” Now every job — a client email, ten meta descriptions, a monthly report — comes back in your voice, with your rules already applied, without you saying any of it again. And before a big change, the agent shows you its plan first: “I'll read the brief, draft 10 descriptions, flag two that need a stat.” You approve, it runs. That's L2: your standards, written once, enforced every time.
CLAUDE.md (your standing rules)
A plain text file the agent reads at the start of every task — your voice, your do's and don'ts, where things get saved. You write it once in normal language; it isn't code.
Why it matters here — It's the difference between briefing a new freelancer every morning and having one who already knows how you work. Write the rule once and it's applied on every run, automatically — no more repeating yourself.
Planning before acting
A mode where the agent proposes what it's going to do — the steps, the files it'll touch — and waits for your nod before it changes anything.
Why it matters here — You catch a wrong turn in one sentence of plan instead of cleaning up after a finished job. On real client work, a 10-second look at the plan is the cheapest safety you'll ever add.
Write your rules down once
Create a CLAUDE.md in your working folder and put your standing rules in plain language — voice, formatting, what to never do, where to save things. Keep it short and real; you can add to it any time the agent does something you'd correct.
Turn on planning
Tell the agent to plan before it acts — to show you its steps and wait for your go before changing anything. You can ask for this by voice at the start of any job; for bigger tasks, make it the default.
“Before you change anything, read CLAUDE.md and show me your plan first. Wait for my go before you touch any files.”
Approve the plan, then let it run
Read the short plan it gives you — does it match the outcome you want? If yes, say go. If not, correct it in one line and let it re-plan. Once it runs, it's already following the rules in your file, so the output comes back in your voice with your standards baked in.
How you know it's working
Before L2: you re-brief the agent from scratch every session — your voice, your rules, your formatting — and it sometimes barrels ahead and does the wrong thing before you can stop it.
- A CLAUDE.md holds your standing rules, and the agent applies them on every task without being reminded.
- The agent shows you a short plan and waits for your go before it changes anything.
- Output comes back in your voice and your format by default — you stopped repeating your preferences.
- You can hand off bigger jobs because you see the plan before the work, not after.
Make it stick. Take the one task you run most and write its rules into CLAUDE.md today — then run the next real one with planning on. Each time you'd have corrected the agent, add that correction as a line. The file becomes your operating standard, working for you on every run.
Our skills for this rung
Linked items are founding-circle skills — clone the repo and run ./install.sh from the skills folder. Unlinked items are practices you build by doing.
Your CLAUDE.md — written rules, planning mode on. Share the file.
Clear this and you've genuinely cleared the rung — not read about it. Keep the proof; it's how you place yourself on the ladder.