New Campaign

You are setting up a new landing page campaign. Read `.claude/commands/CAMPAIGN_LEARNINGS.md` for deeper context and rationale, and apply those learnings throughout this process.

Guide the user through these phases. Move through them naturally — don't number them or say "Phase 1". Just have a conversation.

## Pre-flight: Verify the FULL MCP toolset is connected (ALWAYS the first thing — never skip)

Before asking the marketer any campaign questions, **silently check that EVERY tool the campaign flow will need is available in this Claude Code session**. The pre-flight check exists to prevent the worst failure mode: marketer goes through the whole brief, you call `deploy_landing_page` two minutes later, and the tool isn't actually exposed by the MCP — they have to close VS Code, restart, and start the brief over.

`list_brands` working does NOT mean `deploy_landing_page` works. Partial-connection states are real (e.g. Claude Desktop connector loaded a stale subset of tools, MCP server was redeployed but the client hasn't refreshed, the marketer is on the http connector but only some tools registered). Check the entire inventory.

### How to check

Look at your tool inventory — these **11 tools must all be present** before you start the brief:

| Tool | Used in |
|---|---|
| `list_brands` | Phase 1 brand selection |
| `deploy_landing_page` | Phase 5 deploy (standard signup) |
| `update_landing_page` | Phase 5 iterate (standard signup) |
| `upload_asset` | Phase 3 image swap |
| `teardown_landing_page` | Phase 5 cleanup |
| `upload_dist` | Phase 5 deploy (custom-page / imported HTML) |
| `deploy_custom_page` | Phase 5 deploy (custom-page / imported HTML) |
| `update_custom_page` | Phase 5 iterate (custom-page / imported HTML) |
| `teardown_custom_page` | Phase 5 cleanup (custom-page / imported HTML) |
| `setup_domain` | Phase 5 custom domain |
| `check_ssl_status` | Phase 5 SSL verification |

To verify quickly: call `list_brands` once (it's read-only). If that works, also confirm the other 10 tool **names** are present in your tool inventory. Don't actually call the destructive ones — just check the names.

### If ANY tool is missing — STOP. Don't ask a single campaign question.

"Before we build anything, I need the Campaign Studio MCP fully connected — right now I can see [list which tools you can see, e.g. 'list_brands but not deploy_landing_page or upload_dist']. Without all of them the deploy step at the end won't work, and we don't want to waste your time on a brief we can't ship.

To fix:

1. From the repo root, run: ./scripts/setup.sh
2. If it complains about missing credentials, ask Ziga to DM you the service account JSON file and drop it in mcp-server/credentials/. Re-run setup.sh.
3. Close and reopen VS Code completely. (Claude Code caches the MCP tool list at session start. A reload-window isn't enough — full quit + relaunch.)
4. Reopen this folder, run /new-campaign again. I'll re-check the full inventory.

I won't ask any brief questions until all 11 tools are visible."

Run this check at the **start of every `/new-campaign` invocation**, even within the same session — schemas can refresh, and a single `list_brands` call plus inventory inspection is trivial cost vs. losing a marketer's brief.

### Recovery — if a tool fails MID-FLOW despite passing pre-flight

The pre-flight catches most cases, but tool availability can technically change during a long session (MCP server redeploy, network blip, client-side cache eviction). If you call a tool later and it fails with "tool not found", "transport closed", or similar transport-level error (not a tool-internal error like a bad arg or 4xx response):

1. **Don't retry blindly.** It won't recover within this session.
2. **Stash the marketer's context immediately.** Tell them:

"The MCP just dropped one of the deploy tools — [tool name]. Don't worry, none of the brief is lost. Before I forget anything, here's everything you've told me so far so you can paste it back into a fresh session:

[recap the brief: brand, provider, campaign name, content, variants, formFields, UTM/Klaviyo list, anything custom they asked for]

Now: close VS Code completely, reopen, and run /new-campaign again. When I ask for the brief, paste that recap back. I'll pick up where we left off and re-run the failing step."

3. The recap is the recovery — the marketer doesn't need to re-explain anything.
4. After they restart, re-run pre-flight first. If it fails again, the underlying issue isn't transient — escalate to the developer who set up the MCP.

This recovery flow assumes the MCP was working when pre-flight ran but failed later. If pre-flight didn't actually pass and you skipped it, that's the bug — Phase 0 is mandatory.

## Phase 0: Template offer (ask before collecting the brief)

Before the brief, check for reusable templates.

1. List the contents of `templates/` in the template repo.
2. If there's at least one template folder, show the user the options with their `meta.json` descriptions:
> "I can start this from scratch, or reuse one of these templates: **wntt-quiz-standard** (WNTT quiz layout, 4-result scoring). Want to start from a template?"
3. If they pick one, **skip to Phase 2 Build → Load-from-template path** and only ask for the campaign-specific fields (slug, Klaviyo list id, copy differences, domain).
4. If they say "start from scratch" or there are no templates, continue to Phase 1 normally.

Never block if templates are missing — just proceed.

## Phase 1: Brief

Collect the information you need. Group related questions — don't ask one at a time.

### Subscriber provider (ALWAYS the first question on signup campaigns)

Before brand selection, ASK: **"Are subscribers going to Klaviyo or Beehiiv (HSR)?"**

- **Klaviyo** — DOAC, WNTT, and almost every brand
- **Beehiiv** — HSR (Hot Smart Rich) only

Routing rules — flag and confirm if the answer doesn't match the brand:

| Provi