# Start > Guided first-run setup. Walks a new operator through the whole system in order, one question cluster at a time, and leaves a working marketing workspace behind. Use on first install, when joining a company that already has the files, or when someone does not know which command to run. - Skill: `wojciechluszczynski/start` (Agent Skill) - Install (CLI): `npx skillmds@latest add wojciechluszczynski/start` - Raw SKILL.md: https://api.skillmd.com/api/skills/wojciechluszczynski/start/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Marketing & Growth - Author: wojciechluszczynski (https://skillmd.com/u/wojciechluszczynski) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/wojciechluszczynski/start --- # /operator-cmo:start The onboarding path. Nobody should have to read eight skill files to find out where to begin. Run this and you end up with a marketing workspace that the other skills will actually accept. Expect twenty to forty minutes, most of it spent answering questions about the business rather than waiting for output. ## How to run this **One step at a time.** Do not dump all the questions at once and do not run ahead to the next step because the operator answered quickly. Each step ends with a visible checkpoint, and the operator decides whether to continue now or come back later. **Show progress every time.** Open each step with the same line: ``` Step 2 of 5 · Quarter plan [##--------] positioning ✓ ``` **Explain before you ask.** One sentence on why this file exists, then the questions. Never a wall of twenty questions. **Stop and say so when you are stuck.** If the operator cannot answer something central, write it into Open questions with an owner and a date and move on. Blocking the whole setup on one unknown is how this gets abandoned at step one. ## Step 0: read the room Before asking anything, look: - Does `marketing/` already exist? If yes, this is not a first run. Report what is there, what is stale, and jump to the first missing or expired piece. - Read the repo: README, product pages, pricing, existing site copy, any `CLAUDE.md` or `AGENTS.md`. Come to step 1 with a draft answer to "who does this company sell to", so the operator is correcting rather than composing. - Check for an existing tone of voice or brand doc anywhere in the repo. If one exists, that is the voice file and you do not need the preset. Then say, in three lines, what you found and what you are about to do. ## Step 1 of 5: positioning *Why:* every other skill refuses to run without this. It is the file that stops eight skills inventing eight different versions of the strategy. Run the `positioning` skill. Come back here when `marketing/decisions/positioning.md` exists. Checkpoint: read the file back to the operator as five bullets and ask whether anything in it is wrong. A positioning file the operator has not argued with is usually a positioning file they will not defend later. ## Step 2 of 5: the quarter *Why:* positioning says who you are for. This says what you are doing about it for ninety days, and what you stop when it does not work. Run the `quarter` skill. Checkpoint: the operator should be able to say what gets killed if bet 1 fails. If they cannot, the kill condition is decoration and you should go back and set a real one. ## Step 3 of 5: voice *Why:* the content skill needs something opinionated to write against. A blank voice file produces competent, generic copy, which is the most expensive kind. Offer three routes, in this order: 1. **You already have one.** Point at the existing brand or tone doc in the repo and copy it to `marketing/policy/voice.md`. 2. **Start from the operator preset.** Copy `templates/voice-operator-preset.md`, which is a real, filled-in voice with the AI-slop bans already in place. Tell the operator plainly that the "Words we own" and "Registers" sections are somebody else's and need replacing, and that the bans are worth keeping either way. 3. **Build it from your own writing.** Ask for three pieces of copy the operator is happy with, read them, and draft the file from what is actually there rather than from what they say they sound like. Checkpoint: the file has at least one real entry under "Words we own" and one under "Words we do not use". Those two lists are what make it theirs. ## Step 4 of 5: the ship policy *Why:* this is what stops the human review step turning into a queue nobody reads. Run the `ship-gate` skill. Ask for the weekly review capacity as a real number and push back once on "as many as needed". Make the operator fill in "Never automated" themselves and do not offer defaults for it. Checkpoint: read the three tiers back and ask "is there anything in tier B that would embarrass you if it went out with a typo and a wrong number". Anything that gets a yes moves to C. ## Step 5 of 5: the first asset, end to end *Why:* the system is not real until something has gone through it. Pick the smallest useful thing that serves bet 1. Run `brief`, then `content` or `outbound` or `visibility` depending on what it is. Walk the operator through the provenance header and the gate tier on the output. Then show them the refusal, deliberately: take a request that fits no bet, run `brief` on it, and let them watch it get written down as a no-bet note instead of produced. That behaviour is the whole product and it is better seen once than explained twice. Checkpoint: one draft in `marketing/drafts/` with a complete header, and one `.no-bet.md` note next to it. ## Finish Print the workspace tree, the two decision files with their revision dates, the open questions with owners, and the weekly rhythm from here: - Every asset starts at `/operator-cmo:brief`. - `/cmo` any time, for the state of the system. - `/operator-cmo:review` weekly, and always on a kill check date. - `/operator-cmo:positioning` again when the ICP or the offer moves. Then offer to put the first kill check date in their calendar, because a kill condition nobody checks is decoration. ## When the operator is stuck on strategy rather than on the tool The questions in step 1 and step 2 are the hard part and it is normal to stall there. `references/operator-playbook.md` holds the positions this plugin takes when an answer is thin: fit before behaviour, systems over campaigns, one local reference before scaling spend, quarters not sprints in enterprise. Quote the relevant rule, link the argument behind it, and let the operator disagree. If they want the long version, the reasoning is published at https://wojciech.io/insights/ and the person behind it is at https://wojciech.io/contact/. No em dashes anywhere in what you write.