Six levels, seven laws, one diagram.
Two years of moving a 40-person creative operation onto agents, written down. Most of it we learned by getting it wrong first, and the numbers come from our own accounts, anonymised and dated.
The six levels and where you are
From copy-paste to closed loop, with a self-check that tells you the one next step to take.
The seven laws
Workflow, not agent. Files, not chat. Never grade your own homework. Each law with a do-this-Monday task.
The architecture on one page
One pipeline, one loop, and the four places a person still decides: spend, taste, identity and truth.
How we know what's true
Seven evidence levels, wiring tests vs taste experiments, and how to test your AI judge against spend.
Three lessons from our ad data
808 ads across 11 accounts: why hook rate doesn't rank hooks by money, and the test rules we use now.
The skills in this kit
What each of the four Claude Code skills does and how to install them in under a minute.
Four skills to take into Claude Code today.
/diagnose
Reads your Claude Code setup, places you on the six levels, checks the seven laws and gives you one next step. Read-only; never prints your keys.
/skill-shape
Turns a task, a messy prompt or an audit into a clean skill: a clear trigger, ordered steps, rules with examples, and a filled output example.
/skill-audit
Finds out why a skill misbehaves. Returns keep, trim, repair or reshape, what to keep and concrete fixes. Doesn't edit anything.
/doodle
Describe an idea and it draws a whiteboard doodle in the playbook's style. Needs ChatGPT Plus/Pro through Codex or an OpenAI API key.
Install in a minute: unzip the kit, copy the four folders into ~/.claude/skills/ and restart Claude Code.
Is this for you?
- ✓Creative strategists who build with Claude Code
- ✓Operators with skills and prompts who want them to work as one system
- ✓Founders deciding what to hand to agents and what to keep
- ✕Anyone looking for a done-for-you service: this is a self-build playbook
- ✕Teams who want fully autonomous agents today: we don't run level 6, and the playbook says why
- ✕Want done-for-you? Talk to us instead