---
name: design-funnel
description: Meta-router for design work. Takes a one-line intent, scores its fuzziness, and routes you through the right chain of first-ask → boost-prompt → design-designer. Always confirms routing before executing. Auto-saves each step's output to the daily note. Design-specific only. Trigger with "/design-funnel", "help me design X", "I want to design something".
---

# Design Funnel

A meta-skill that routes design work through the right chain of interview skills. You give a one-line intent, funnel picks the chain, you confirm, it runs each step in order and passes context forward so you never answer the same question twice.

## When to Use

Trigger when the user says:
- "/design-funnel"
- "help me design X"
- "I want to design something"
- "walk me through a design prompt"
- "build me a landing page / deck / prototype" — but they haven't started the interview yet

Do NOT trigger when the user is already mid-interview in first-ask, boost-prompt, or design-designer. Funnel is for the start of the funnel only.

## Scope Lock

Design-specific only. The target is always claude.ai/design at the end of the chain. If the user's intent isn't a design artifact (landing page, deck, prototype, marketing site, scroll-site, dashboard, app UI), tell them funnel isn't the right tool and suggest `/boost-prompt` or `/first-ask` standalone.

---

## How It Works

### Phase 1 — Capture Intent

Ask exactly one question:

> **In one sentence, what do you want to design?**

Free text. Don't pre-filter or pattern-match the type yet. Just capture.

### Phase 2 — Score Fuzziness

Score the user's one-liner on 3 axes. Each axis is binary: clear (1) or fuzzy (0).

| Axis | Clear if... | Fuzzy if... |
|---|---|---|
| **Audience** | Names a specific group, segment, or persona ("Honeybird wholesale sellers", "solo founders who hate noise") | Generic or missing ("for people", "for users", "for everyone") |
| **Offer / value** | Names the concrete product, service, or outcome ("a 15-min underwrite tool", "our Q2 investor pitch") | Vague noun ("something cool", "a tool", "a thing") |
| **Format** | Names a specific artifact type ("one-page landing with pricing", "7-slide deck", "mobile app prototype") | Missing or hedged ("a website maybe", "some kind of page") |

Total the score (0-3).

### Phase 3 — Pick Chain

| Score | Chain | Why |
|---|---|---|
| 0-1 | **Chain 1**: first-ask → boost-prompt → design-designer | Intent is fuzzy enough that first-ask needs to force audience / offer / format clarity before any prompt work |
| 2 | **Chain 2**: boost-prompt → design-designer | Intent is clear but the prompt shape isn't. boost-prompt structures, design-designer targets |
| 3 | **Chain 3**: design-designer only | Intent is crisp. Go direct |

### Phase 4 — Confirm Routing (ALWAYS — no kill switch)

Show the user:
- Their one-line intent (verbatim)
- The score on each of the 3 axes with a one-sentence reason
- The chain that was picked
- Offer override: "If you want a different chain, say which"

Wait for confirmation. Accept "go", "yes", "run it", "confirmed" as the confirm signal. Accept "chain 1 / 2 / 3" as an override. Do NOT proceed without explicit confirmation.

### Phase 5 — Execute the Chain

Run each skill in order. Between each step:
1. Show the output of the completed step in a single block
2. Append it to `daily/YYYY-MM-DD.md` under a header `## Design Funnel — {intent slug} — Step N: {skill name}`
3. Pass the output as context into the next skill's interview so the user does NOT re-answer the same questions

**Chain 1 execution order**:
1. Run `first-ask` with the one-line intent as the starting prompt. first-ask will ask its usual questions and output a plan.
2. Feed first-ask's plan into `boost-prompt` as the "rough draft". boost-prompt polishes.
3. Feed boost-prompt's output into `design-designer` via its "skip the interview" quick path.
4. Design-designer produces the final claude.ai/design-ready prompt.

**Chain 2 execution order**:
1. Run `boost-prompt` with the one-line intent as the rough draft.
2. Feed output into `design-designer` quick path.
3. Design-designer produces final prompt.

**Chain 3 execution order**:
1. Run `design-designer` directly with the full 10-question interview. The one-line intent becomes Q2 seed.

### Phase 6 — Final Delivery

When the chain completes, output:
1. The final Claude Design-ready prompt in a copy-paste block
2. The 3-point briefing from design-designer (what to attach, what it'll still ask, next move)
3. A line confirming the full trail was saved to today's daily note with the intent slug so the user can find it later

Close with: **"Paste the prompt above into claude.ai/design. Full trail saved to `daily/YYYY-MM-DD.md` under `Design Funnel — {slug}`."**

---

## Auto-Save Behavior

Every step writes to `daily/YYYY-MM-DD.md` automatically. No asking. Format:

```markdown
## Design Funnel — {intent-slug} — {timestamp}

**Intent**: {one-line intent verbatim}
**Score**: audience={0|1} offer={0|1} format={0|1} → total {N} → {Chain name}

### Step 1: {skill name}
{output}

### Step 2: {skill name}
{output}

### Step 3: design-designer
{final prompt}

### Briefing
{3-point briefing from design-designer}
```

Slug format: kebab-case, first 5 meaningful words of intent. Example: "landing page for honeybird wholesale sellers" → `landing-page-honeybird-wholesale-sellers`.

If the daily note doesn't exist, create it. If it exists, append. Never overwrite.

---

## Routing Edge Cases

- **User overrides to a lighter chain**: If scored at 0 but user says "just run chain 3", respect it but warn once: "Heads up — your intent is missing {audience/offer/format}. design-designer will ask more questions than usual. Okay to proceed?"
- **User gives a non-design intent**: If the one-liner is clearly not a design artifact ("write me an email", "fix this bug"), respond: "Funnel is design-specific. Try `/boost-prompt` for a prompt-polish or `/first-ask` for a generic task interrogation." Exit.
- **User abandons mid-chain**: If the user says "stop" or "never mind" between steps, save what you have so far to the daily note with a `### Status: abandoned at Step N` line. Don't delete partial work.
- **User wants to rerun a single step**: If they say "redo step 2" or similar, keep all prior steps, rerun only the named step, replace that step's block in the daily note.

---

## What Funnel Does NOT Do

- Does not pick the chain silently. Always shows the score and waits for confirmation.
- Does not route to anything other than first-ask, boost-prompt, design-designer.
- Does not handle non-design targets. No Notion pages, no code, no emails.
- Does not skip the daily note save. That's non-negotiable for session continuity.
- Does not collapse the chain into a single mega-prompt. Each skill runs its own interview; funnel is the conductor, not a replacement.

---

## Example Run (Chain 2)

User: `/design-funnel`

Funnel: "In one sentence, what do you want to design?"

User: "a one-page landing for Honeybird's new cash-offer tool, aimed at sellers who are out of state"

Funnel:
> **Intent**: "a one-page landing for Honeybird's new cash-offer tool, aimed at sellers who are out of state"
>
> **Score**:
> - Audience = 1 (out-of-state Honeybird sellers — specific)
> - Offer = 1 (cash-offer tool — concrete)
> - Format = 1 (one-page landing — specific)
>
> Total: 3 → **Chain 3: design-designer direct**
>
> Confirm to run, or say "chain 1" / "chain 2" to override.

User: "go"

Funnel runs design-designer. design-designer asks its 10 questions using the intent as seed for Q1 and Q2. User answers. design-designer outputs final prompt. Funnel appends everything to today's daily note and closes.