mktg-setup

name	mktg-setup
description	First-run conversational setup wizard for marketing-cli. Use when /cmo detects no populated brand files and no cmo-preferences.md, when the user just installed marketing-cli for the first time, or when they say things like "set up marketing", "start marketing", "marketing wizard", "I just installed marketing-cli", "onboard me", "first time", "initial setup". Asks 3 sharp questions (identity, posture, distribution preferences), records the answers to brand/cmo-preferences.md, then hands off to /cmo's foundation flow which spawns 3 research agents in parallel. The user never has to type /mktg-setup directly — /cmo auto-routes here on first run.
allowed-tools	["Bash(mktg *)","Bash(mkdir *)","Bash(test *)","Bash(cat *)"]

/mktg-setup — First-Run Conversational Setup

When This Runs

/cmo invokes this skill automatically when both conditions are true:

1. brand/cmo-preferences.md does not exist (or is template content), AND
2. mktg status --json reports brandSummary.populated < 3.

The user typically never types /mktg-setup. They install marketing-cli (npm i -g marketing-cli), open Claude Code in any project directory, type /cmo (or just say "help me with marketing"), and /cmo's activation routes here. The wizard is a conversational on-ramp, not a slash command users have to memorize.

What This Skill Does

Three AskUserQuestion calls. That is the entire wizard. No batched questions, no progress bars, no long forms.

Each answer becomes a persistent decision in brand/cmo-preferences.md. After question 3, hand off to /cmo's foundation flow (3 research agents in parallel — see cmo/rules/sub-agents.md).

The wizard's job is to record the small set of decisions Claude cannot make for the user. Filling out the 10 brand files is /cmo's job, not the wizard's.

On Activation

1. Run mktg status --json --fields project,brand,brandSummary from the project root.
2. If brandSummary.populated >= 3 AND cmo-preferences.md doesn't exist: a returning user got upgraded to a version that introduced preferences. Tell them: "I see you have a populated brand but no preferences file yet. I'll write one based on what's already here." Skip directly to step 3 (write preferences) using inferred defaults from existing brand files.
3. If cmo-preferences.md already exists and is non-template: skip the wizard entirely. Hand back to /cmo immediately. Tell the user: "Looks like you've already onboarded. Hand off to /cmo for your next move."
4. Otherwise (genuine first run): proceed to Step 1.

---

Step 1 — Ask: Identity

Step 1 is a single-select question (1a) plus exactly one free-text follow-up (1b or 1c). Don't merge them and don't add extra prompts in between.

Step 1a — Setup mode

Use AskUserQuestion. Single-select.

Question text: "How should we set up your brand?"

Options:

• "From a URL — I'll scrape your site to seed brand files (recommended if you have a website)"
• "Local conversational setup — I'll ask a few questions instead"
• "Pre-launch / open source / personal brand — describe in 1 sentence"

The user's choice determines the follow-up:

• "From a URL" → go to Step 1b (URL capture).
• "Local conversational setup" OR "Pre-launch / open source / personal brand" → go to Step 1c (sentence capture).

Step 1b — URL capture (only if Step 1a = "From a URL")

Use AskUserQuestion. Single-select with the "Other" free-text path (so the user can paste the URL inline).

Question text: "Paste your URL"

Options:

• "I'll paste the URL in the next message" — the user types the URL in their next chat message.
• (Use the built-in "Other" free-text option for inline URL entry — paste-and-go.)

Capture the URL exactly as the user provides it. Do NOT validate it inside the wizard — no regex, no protocol check, no normalization. Pass the raw string straight to the CLI:

mktg init --from <url>

mktg init --from is the single source of truth for URL handling. If the URL is malformed, unreachable, or scrape-blocked, the CLI surfaces the error; the wizard relays that error to the user and falls back to the local sentence path (offer "I'll describe it in one sentence instead" without re-asking Step 1a).

On success, the scrape seeds brand/voice-profile.md, brand/positioning.md, brand/audience.md, and brand/competitors.md. Hold the captured URL and the "scrape ran" status (yes/no) for Step 5.

Step 1c — Sentence capture (only if Step 1a = "Local conversational setup" OR "Pre-launch / open source / personal brand")

No URL here — this is the deliberate no-URL path. Ask a single short follow-up:

"Got it — describe what we're marketing in one sentence."

Capture the response verbatim. Hold it as the seed for brand/voice-profile.md (it lands in the "What we are" section during Step 5 / foundation flow). Don't run mktg init --from; the foundation research agents fill the gaps.

---

Step 2 — Ask: Posture

Use AskUserQuestion. Single-select.

Question text: "What's the marketing posture you want?"

Options:

• "Aggressive launch" — Product Hunt + Hacker News + X campaign blitz. Best for new product launches with momentum to capture.
• "Steady authority build" — SEO + newsletter + slow-burn thought leadership. Best for long-game category ownership.
• "Founder-led / personal brand" — voice-driven content, personality forward. Best when the founder IS the differentiator.
• "Product-led growth" — free tier hooks, referrals, viral mechanics. Best when the product itself drives acquisition.

This is the single most important answer. It shapes which skills /cmo prioritizes downstream:

Posture	Skills /cmo prioritizes
aggressive-launch	launch-strategy, startup-launcher, social-campaign, direct-response-copy
steady-authority	seo-content, keyword-research, newsletter, ai-seo
founder-led	brand-voice, voice-extraction, content-atomizer, mktg-x
product-led-growth	free-tool-strategy, referral-program, lead-magnet, conversion-flow-cro

---

Step 3 — Ask: Distribution

Use AskUserQuestion. multiSelect = true.

Question text: "Which distribution channels should we wire up first?"

Options:

• "Typefully" — X/Twitter and threads. Best for short-form and thread-first audiences.
• "Postiz" — LinkedIn, Reddit, Bluesky, Mastodon, Threads, Instagram, TikTok, etc. (30+ providers via the postiz catalog).
• "Resend" — transactional email and newsletter delivery.
• "Skip for now — I'll connect later" — /cmo will use --adapter file for export until the user wires something up.

Multiple selections are fine. If the user picks "Skip", record that explicitly — a deliberate skip is signal, not absence of data.

---

Step 4 — Ask: Studio (beta)

Use AskUserQuestion. Single-select.

Question text: "Want to use the Studio dashboard? (beta)"

Options:

• "Yes — auto-open the Studio at the end of setup (recommended)" — /cmo will run mktg studio automatically at the end of foundation flow and on subsequent activations when the user says "show me the dashboard."
• "No — keep me in the CLI only" — /cmo will not auto-launch Studio. The user can still run mktg studio manually any time.

Why this question: Studio is in beta. Some users prefer pure CLI workflows (especially headless/agent-only environments). Recording the choice once means /cmo never nags later.

If the user picks "No", /cmo reads studio_enabled: no from preferences and skips the studio launch step in its foundation flow. If they later change their mind, they can edit brand/cmo-preferences.md directly or just say "let's open studio" and /cmo will respect the explicit ask.

---

Step 5 — Write Preferences

Create brand/cmo-preferences.md with this shape:

# CMO Preferences

> Recorded by /mktg-setup on {ISO 8601 date}. Edit this file to change /cmo's runtime behavior.
> /cmo reads this file on every activation and uses it to shape skill prioritization and distribution routing.

## Identity

- **What we're marketing**: {URL or one-sentence description from Step 1}
- **Setup mode**: {from-url | local | pre-launch}
- **URL**: {captured URL if Setup mode is from-url, otherwise omit this line}
- **Scrape ran**: {yes if mktg init --from succeeded, no if from-url failed or mode is local/pre-launch}

## Posture

- **Mode**: {aggressive-launch | steady-authority | founder-led | product-led-growth}
- **Why this mode**: {one-line rationale based on the user's choice — pull from the option description in Step 2}
- **Skill priority shortlist**: {comma-separated list from the posture-to-skills mapping}

## Distribution

- **Selected**: {comma-separated list of selected platforms, or "none" if skipped}
- **Skipped**: {whichever options weren't selected}
- **Default adapter**: {typefully if X selected and only X | postiz if multi-platform | file if skipped}

## Studio

- **studio_enabled**: {yes | no}
- **Beta**: yes (Studio is in beta — record the user's preference; /cmo respects this on every activation)
- **Auto-open at end of setup**: {yes if studio_enabled is yes, no otherwise}

## Notes

- This file is the persistent contract. /cmo reads it on every activation.
- To change preferences, edit this file directly OR ask Claude to "re-run mktg setup" (it will overwrite this file).

Make sure the directory exists before writing:

test -d brand || mkdir -p brand

---

Step 6 — Hand Off to /cmo

SEO backend handoff (before spawning agents): run mktg catalog info openseo --json --fields configured,missing_envs. OpenSEO is the default SEO data plane (measured KD, volume, SERP, backlinks, GSC).

• configured: true → tell the user: "OpenSEO is connected. After foundation, I'll link a project (mktg seo link-project) so keyword research uses measured metrics." After the foundation chain, if .seo/openseo.json is still missing, recommend /openseo-project-setup as the first SEO move.
• configured: false → one line only, no blocking: "OpenSEO isn't set up — keyword metrics will be unknown until it is. Setup takes a minute: openseo.so → connect the MCP, or set OPENSEO_API_KEY." Never install anything or ask for keys inside the wizard.

Tell the user clearly what's happening next. Tailor the closing line to their Studio choice:

• studio_enabled = yes → "Recorded your preferences. Now I'm spawning 3 research agents in parallel — brand voice, audience, and competitors — to fill out your foundation. ~5 minutes. Then I'll auto-open the Studio dashboard."
• studio_enabled = no → "Recorded your preferences. Now I'm spawning 3 research agents in parallel — brand voice, audience, and competitors — to fill out your foundation. ~5 minutes. Studio stays off per your choice; I'll surface results in the CLI."

Then invoke the foundation flow per cmo/rules/sub-agents.md:

1. Spawn mktg-brand-researcher with the identity context from Step 1
2. Spawn mktg-audience-researcher with the posture context from Step 2
3. Spawn mktg-competitive-scanner with both

Wait for all 3, then chain:

• /landscape-scan for ecosystem grounding
• /positioning-angles to lock the angle
• If studio_enabled is yes: run mktg studio to auto-open the dashboard with the populated state.
• If studio_enabled is no: skip the Studio launch. Print a one-line summary of what was populated and the next recommended skill in the CLI. Tell the user: "Studio's off per your preference. If you change your mind, edit brand/cmo-preferences.md or just say 'open studio'."

---

Anti-Patterns

Anti-pattern	Instead	Why
Asking more than 3 questions in the wizard	Stop at 3 — hand off to /cmo for the rest	More questions = abandonment. /cmo's research agents fill the rest in 5 minutes via real research.
Asking for a URL before letting the user pick a setup mode	Always ask Step 1a (setup mode) first, then branch to Step 1b (URL) or Step 1c (sentence)	URL is just one of three paths. Pre-launch / open-source / personal-brand users don't have a URL yet, and the local conversational path bypasses the scrape entirely. Forcing a URL up front blocks them.
Validating the URL inside the wizard (regex, http vs https, "is this real?")	Pass the raw string straight to mktg init --from <url> and let the CLI handle malformed inputs	The CLI already returns structured errors for bad URLs. Re-validating in the wizard duplicates logic, drifts from the CLI, and frustrates users who paste valid-but-unusual URLs.
Silently routing to brand-voice / audience-research before recording preferences	Always record cmo-preferences.md FIRST	Without it, every future /cmo activation re-runs the wizard. The preferences file is the contract that prevents that.
Skipping the wizard if voice-profile.md exists but cmo-preferences.md doesn't	Run the wizard regardless of other brand files (use the migration path in step 2)	cmo-preferences.md is the persistent contract — separate from brand content. A user who upgraded from pre-mktg-setup needs to record their posture explicitly.
Treating "Skip distribution" as missing data	Record the skip explicitly as selected: none and default_adapter: file	A deliberate skip is signal. The user told us not to wire anything yet — respect that.
Spawning research agents BEFORE writing cmo-preferences.md	Always write the file first	If the user kills the conversation mid-research, we lose the preferences. The file write is the savepoint.
Letting the user re-run the wizard accidentally	After writing the file, /cmo on the next activation will see it and skip the wizard	This is the whole point of the persistent contract. /cmo only re-routes here when the file is missing.

---

Iteration / Re-Running

The user can change their preferences three ways:

1. Edit brand/cmo-preferences.md directly — it's just markdown.
2. Ask Claude: "I want to change my marketing posture" — /cmo will re-route here, overwriting the file.
3. Future: mktg setup --force (CLI flag, not yet implemented at v0.5.0; tracked for a future release).

---

Dependencies

• Reads: mktg status --json, optionally URL via mktg init --from <url>.
• Writes: brand/cmo-preferences.md (new). Optionally seeds brand/voice-profile.md if the user provided a sentence in Step 1.
• Hands off to: /cmo foundation flow (which spawns the 3 research agents and chains /landscape-scan → /positioning-angles → mktg studio).