One command.
A full client launch.
/goal

---
name: goal-loop
description: Run a multi-deliverable goal to completion. Use when the user
  gives a list of things to produce for a client or project.
---

When the user gives you a goal with multiple deliverables:

1. SCOPE FIRST. Before building anything, ask up to 4 sharp
   multiple-choice questions covering the decisions that change the
   work (audience, brand direction, format, constraints). Never ask
   open-ended questions you could answer yourself from their website.

2. PLAN IN DELIVERABLES. Restate the goal as a checklist. Every item
   gets a one-line acceptance test, e.g.:
   - landing page → "deployed to a preview URL and screenshot-verified
     on desktop AND mobile"
   - ad creative → "second review pass scores it 8/10 or better
     against the brand doc"
   - document → "formatted, in the client folder, link surfaced"
   Keep the checklist visible and updated as you work.

3. BUILD THE SOURCE OF TRUTH FIRST. Scrape the client's real website.
   Write a brand/facts file with every verifiable number, name and
   claim. Rule: no deliverable may contain a number that is not in
   this file. Never invent testimonials, stats or credentials.

4. WORK THE LIST. One item at a time or in parallel where independent.
   After each item, run its acceptance test yourself. If it fails,
   fix and re-test before moving on. Maximum 2 fix passes, then flag
   the gap loudly instead of shipping quietly broken work.

5. ONLY STOP WHEN THE LIST IS DONE. The final message is a clean
   summary: every deliverable, every link, anything flagged. No
   "next steps I could do" padding. If a deliverable was impossible,
   say exactly why and what is needed.

---
name: brand-dna-builder
description: Build a brand DNA file from a client's real website. Use
  before any creative or strategy work for a new client.
---

Input: the client's website URL (plus any subpages the user names).

1. FETCH the homepage and the 2-3 most important subpages (about,
   services, pricing). Read what is actually there, not what you
   assume.

2. WRITE brand-dna.md with these sections:

   ## Verified proof stack
   A table of every claim you can quote from their site: years in
   business, client counts, results, accreditations, guarantees,
   founder names, phone, email, hours. Mark the source page of each.
   RULE FOR ALL FUTURE WORK: only these numbers may appear in copy.
   If it is not in this table, it does not exist.

   ## Current brand state
   What their site looks and sounds like today: colors (grab real hex
   values), fonts, tone, weaknesses. Be honest.

   ## Direction
   Positioning line (one sentence a customer would repeat), palette
   (5-6 named colors with hex + usage), typography (display + body),
   photography rules (real photos only? what mood?).

   ## Voice
   5 rules with examples. Include a banned-words list (start with:
   unlock, elevate, seamless, hassle-free, game-changer, journey)
   and one sentence the brand would NEVER say.

   ## The one-sentence test
   Describe the exact customer, then: "If [that person] reads the
   page and thinks [desired reaction], it works."

3. Keep it under 2 pages. A brand doc nobody reads protects nobody.

---
name: dr-copywriter
description: Write direct-response copy (ads, landing pages, emails)
  that survives an AI-tell audit. Requires a brand-dna.md.
---

Before writing: read brand-dna.md. Every number you use must come
from its verified proof stack. No exceptions, no rounding up.

WRITING RULES
- Every promise carries a number. "Faster" is not a claim;
  "answered within the hour" is.
- Short sentences. One idea each.
- No em-dashes anywhere.
- Banned words: unlock, elevate, seamless, hassle-free, game-changer,
  revolutionize, "imagine a world", "in today's fast-paced".
- One CTA per piece. Low-commitment beats high-commitment for cold
  traffic (an estimate or a checklist beats "book a call").
- Headlines: write 2-3 options (A/B/C), body copy single version.

SELF-TEST LOOP (mandatory, before delivering)
Score your draft 1-10 on each, rewrite until every score is 9+:
1. Read-aloud test: would the business owner actually SAY this
   sentence to a customer, on the phone, without cringing?
2. Specificity: does every claim carry a verifiable number or name?
3. AI-tell scan: any banned words, em-dashes, three-adjective
   stacks ("fast, reliable, and professional"), or "it's not X,
   it's Y" constructions? Kill them.
4. One-big-idea: can you state the piece's single idea in one
   sentence? If it has two ideas, split it.
5. The customer test from brand-dna.md.

Report the final scores with the delivery. If any dimension is
below 9 after 3 rewrites, deliver anyway and flag which test failed
and why. Never silently ship a failing draft.

---
name: ad-statics-designer
description: Produce Meta ad statics as HTML/CSS rendered to PNG.
  Real text, real brand fonts, real photos. Requires brand-dna.md
  and Playwright (npm i playwright).
---

CORE RULE: image models misspell and hallucinate text. So the ad is
an HTML page: brand fonts loaded from Google Fonts, real copy as
real text, the client's actual photo as the background. Then
screenshot it. Nothing is ever "drawn".

For each ad concept:
1. Write the on-image copy first (use dr-copywriter rules).
   Max 20 words on the image. One loud message. One solid CTA button.
2. Build a single HTML file: a fixed-size div (1080x1350 for 4:5,
   1080x1080 for 1:1), brand colors from brand-dna.md, the photo as
   a full-bleed img with a gradient scrim where text sits, logo
   visible (top corner chip), headline at 72px+ equivalent.
3. Screenshot with Playwright:
   const { chromium } = require('playwright');
   const b = await chromium.launch();
   const p = await b.newPage({ viewport: {width:1080, height:1350},
     deviceScaleFactor: 2 });
   await p.setContent(html, { waitUntil: 'networkidle' });
   await p.screenshot({ path: 'ad-01.png' });
4. LOOK at the rendered PNG (read it back). Check: text legible on
   the photo? headline big enough at thumbnail size? logo crisp?
   copy verbatim? Fix and re-render until it passes.

BATCH RULE: a set of 4 ads should MIX patterns — one photo hero,
one pure-typography statement on a flat background, one offer card
with 3 checkmarks, one big-number comparison. Four lookalike heroes
is one ad tested four times.

---
name: video-ad-scriptwriter
description: Write 25-35s avatar-less video ad scripts (b-roll +
  kinetic text + voiceover) for AI video tools. Requires brand-dna.md.
---

THE MATH THAT MAKES OR BREAKS IT
Voiceover word budget = duration x 2.3 words/second.
30s video = 69 words MAX. Count them. Over budget = the ending gets
cut off or rushed. Under budget is fine (b-roll fills the space).

For each script, produce:

1. VO (55-75 words for 30s). Conversational, hook inside the first
   2 seconds, ends with a low-commitment CTA ("free estimate in 60
   seconds" beats "contact us today"). Every number from the
   verified proof stack.

2. SCENES: 6 timed directions, each ending with a boxed text
   overlay of 3-6 words. Format:
   [0-3s] HOOK: .
     Boxed overlay (top): ""
   [3-8s] ... overlay: ""
   ...
   [25-29s] solid brand-color end card. Centered boxed overlay:
     ""

3. RULES baked in:
   - describe visuals concretely ("close-up of a phone banking app
     as a payment lands"), never abstractly ("success imagery")
   - specify the casting and setting to match the market, or the
     tool defaults to generic global stock
   - brand colors stated explicitly (background + accent hex)
   - one idea per scene; the overlays alone must tell the story
     with the sound off

Write 3 script variants per campaign: 3 different hooks (the
promise, the number, the pain), same offer and CTA.