Creative strategy in Claude Code.

Everything in the core flow runs in Claude Code, in the workspace you built on Day 1. That is the part everyone in the Accelerator has, and it is enough to run your whole creative process.

If you are in Genesis, you also have the Creative OS (the Day 2 Creative Strategy Skill File, the /cs modes, and the bots). That sits on top of this as an accelerator, not a replacement. Wherever it speeds a step up, there is a Genesis only box. No Genesis? Skip those boxes. Nothing in the core flow needs them.

START: feed it before you write

Before a single hook, give Claude Code the three things it needs. This is the "give it X, Y, Z first" question:

Paste this to open a session:

New writing session for {client}.
Read clients/{client}.md, brand/voice.md, and the swipes in swipes/{mechanism}/.
Here is the angle I want to chase: [one line].
Do not write yet. Tell me what you have and what is missing.

DURING: save vs promote

This is the question you asked, and it has a clean answer. The workspace splits in two:

• Workbench holds the work: copy/, creatives/. Today's drafts live here. This is save.
• Memory holds the lesson: winners/, losers/, brand/, clients/, workflows/. This is promote.

The rule of the second time: once is noise, the second time is signal, the third time it should already be a saved play. Promote the lesson that should change your NEXT ad, not every line from today.

END: log it so next time starts ahead

Two commands, every working session. This is the "run THIS at the end every time" question:

Claude Code the session loop (everyone has these)

Claude Code working and verification (everyone has these)

Genesis only: the Creative OS modes

These run inside your Creative OS (the Genesis Creative Strategy tool). They are the upstream thinking: generate here, then bring the chosen output back into your Claude Code workspace to write it in your voice.

Not sure where you are? Type /creative-strategist with nothing after it and it routes you.

1. Load the desk

New writing session for ZenMag.
Read clients/zenmag.md, brand/voice.md, and the swipes in swipes/surprising-culprit/.
Do not write yet. Tell me what you have on this brand and what is missing.

Claude reads your files: it has the voice rules and two matching swipes, it is missing your proof. You point it at the proof line in the client file.

2. Ideate from your material, not a blank page

Give me 10 angles for ZenMag, grounded in the swipes you just read and the
research in research/. No blank-page guesses. Lead with curiosity, not pain.

You pick one: the Hidden Culprit. "It is not stress keeping you up at 3am. Stress burns through a mineral your body needs to stay asleep."

3. Aim it

For that angle, who is the sharpest segment and what awareness level are they?
Use the research, do not invent an audience.

You land on tired-but-wired professionals, 35 to 50, problem-aware.

4. Hooks first, stress-tested

Write 5 hooks for this angle and segment.
Run Luke's Vicious test on each one and tell me which survive.

Claude writes five, scores them, and kills the weak ones out loud: "'There is a mineral that helps you sleep' fails, weakest read is 'a thing helps a thing.' Rewrite: 'Your 3am wake-up is a mineral problem, not a stress problem.'"

5. The script, only after the hook survives

Take the winning hook and write the full short-form video script.
Hook, then payoff, then mechanism, then offer. Keep it caveman simple.

6. Check it before it ships

/adversarial-review

Then the simplification gate:

Rewrite any dense section. Same ideas, same order, simple words.
Swap Latin-root words for Germanic ones.

7. Bank the lesson

/reflect

Claude proposes: promote the winning hook structure to winners/ with the WHY, note the dead "mineral helps you sleep" line in losers/, and add "always name a specific time anchor (3am) in sleep hooks" as a voice rule. You approve. Next ZenMag session, Claude already knows all of it.