Skai Product Launch Plus White Paper

SKAI Product Launch + White Paper

One playbook, six phases, each with a hard gate. Built and proven on the ClaimFlow launch (July 2026). You are the orchestrator: run research in parallel, write the assets yourself, verify every gate before advancing.

All of Hormozi's source material is attached in references/hormozi/ ($100M Offers, $100M Leads, $100M Money Models, $100M Lost Chapters — full PDFs) with pre-distilled framework extractions in references/hormozi-distilled/ (JSON: core frameworks + whitepaper/email/ad/content hooks per book, each mapped to B2B SaaS launch application). Read the distilled JSONs first — only go to the PDFs when a framework needs deeper treatment or the product context differs enough that the distillation doesn't transfer.

Phase 0 — Brief (inputs to lock before anything runs)

• Product + URL, ICP (one sentence, avatar-specific), and the wedge sentence (the "compliance is the wedge, attribution is the retention"-style one-liner; if none exists, write it first — every asset hangs off it).
• The authority concept for the whitepaper: a genuinely novel or contrarian POV (ClaimFlow example: Poincaré/hyperbolic embeddings applied to attribution). No concept → the whitepaper is a commodity PDF; push back per Cardinal Rule 0.
• Current-events domain for newsjacking (regulatory calendar, enforcement actions, platform changes).
• Honesty boundary: which claims describe shipped software vs. roadmap. This gets its own research agent in Phase 1.
• Budget scale for ads, and brand tokens (colors, display/body/mono fonts, and where the .woff2 files live).

Phase 1 — Research fan-out (one Workflow, parallel agents)

Run as a single Workflow with parallel(); sonnet-tier agents; structured-output schemas. Streams:

1. Hormozi frameworks — read references/hormozi-distilled/*.json; escalate to the PDFs (Read with pages, ≤15/chunk) only if needed. Output: frameworks + hooks mapped to THIS product.
2. Current events — WebSearch/WebFetch, 6–12 dated events with primary-source URLs (regulator sites + trade press), each with "why the ICP cares" + a newsjack angle tied to the wedge sentence. Accuracy is mandatory; these anchor the whitepaper.
3. Scholarship + prior art — verify every citation (authors, year, title, venue, URL). Run ≥6 differently-phrased prior-art searches. Output the exact defensible novelty phrasing: "To our knowledge, the first X — the nearest adjacent work is [cite], which does Y not X." Never claim absolute priority; always name the nearest neighbor. Flag any citation the agent couldn't independently verify and drop or replace it.
4. Competitor Meta ads — LIVE, via the user's browser. Known failure mode: Meta 403-blocks ALL automated access to the Ad Library (WebFetch, Apify rendering browsers, headless anything). Do NOT burn agent budget on it. Use claude-in-chrome: open facebook.com/ads/library in the user's real Chrome, type each brand into the search box, click the advertiser Page in the typeahead (keyword search is junk — always select the Page), then get_page_text for the full ad inventory. Record per competitor: active-ad count, verbatim themes/hooks, formats, CTAs. The deliverable is whitespace: message lanes with observed-zero coverage.
5. Honesty pass — an agent reads the actual codebase(s) and returns honest_claims: what a public asset CAN truthfully claim and what it MUST NOT claim. This list is binding on every later phase.

Gate: all streams returned; research saved to /marketing/research/*.json; novelty phrasing locked; honest-claims list in hand.

Phase 2 — The whitepaper

Structure (the persuasion spine is Hormozi's Value Equation; the skeleton is a problems→solutions list):

1. Title by M-A-G-I-C (magnet/avatar/goal/interval/container) with an authority-signal main title.
2. Executive summary — wedge sentence, the two arguments (technical + practical), the novelty claim.
3. "Why now" — the newsjack table: date | event | what it means for the reader. Calendar first, math later.
4. Status quo + where it breaks — name what ships today honestly, then the structural critique.
5. The concept's lineage — history + verified citations; include ≥1 current-year citation proving the concept is a live research area (the "tied to AI current events" beat).
6. The application case — why this concept, this industry; include the hedged novelty claim with nearest-neighbor citation.
7. "What we built / what we're building" — bounded by the honest-claims list; include an explicit what-we-will-not-claim paragraph (radical verifiability IS the sales pitch for skeptical buyers).
8. Why this market — the Starving Crowd four checks (pain/purchasing power/targetable/growing) argued with the research.
9. Self-audit section — 5 diagnostic questions (reveal-a-problem lead-magnet mechanics).
10. Close = stacked offer: named bonuses with jobs ("built for your compliance officer"), a real guarantee, honest capacity-based scarcity, one CTA. Then full references + a fine-print integrity footer (sample-data disclosure, novelty hedge, forward-looking labels).

Voice: plain language, lead with the concrete, no banned-word hype, explain any math in analogies. ~4,000–5,500 words.

PDF pipeline (adapt references/pipelines/build-pdf.mjs): npx marked --gfm → wrap in print-CSS shell with brand fonts embedded base64 from the token woff2s → chrome.exe --headless=new --no-pdf-header-footer --print-to-pdf. Verify typography by screenshotting the HTML (--screenshot) and Reading the PNG — don't ship unverified.

Gate: markdown + branded PDF exist; every citation verified; every claim inside the honesty boundary; screenshot check passed.

Phase 3 — Email plan

• Mechanism before copy: capture page gating the PDF (email + 2–3 qualifying fields = the friction dial), server-side email verification at submit (Hunter — Cardinal Rule 7 is absolute: nothing unverified enters any sending platform), push to Kit/ConvertKit with tags, dedicated sending subdomain with SPF/DKIM/DMARC. Deliver the PDF on-page too — the gate is a toll, not a hostage.
• Sequence (Day 0,1,3,5,8,11,14,18,22,26): deliver-everything → one-small-action → story-with-stakes → the calendar/deadlines → reframe ("vendor handles that" ≠ answer) → the concept in 60 seconds → first ask (pilot + guarantee, menu CTA) → unselling/objections → honest capacity status → 9-word re-engage. Give:ask ≥ 4:1. All urgency real (regulatory dates, true capacity) — never manufactured.
• Post-sequence: monthly pure-value calendar email + two-sided referral ask (dense-network niches compound referrals faster than ads).
• Money-model staging (from $100M Money Models): whitepaper = attraction → pilot = upsell → feature-light tier = downsell → subscription = continuity. Sell only the next step in each email. Consider Founding-cohort lifetime rate + real one-time setup fee (Lost Chapters continuity mechanics).
• Cold outreach is a separate lane and never enters the warm ESP list.

Gate: EMAIL-PLAN.md with full copy for all sends + measurement targets; capture-page build task created (it blocks ads).

Phase 4 — Meta ads (whitespace report + Andromeda design)

Whitespace report from Phase 1's live observations: table of competitor | active count | verbatim themes | formats/CTAs, then the whitespace list (angles with observed-zero coverage), then budget: small-launch default $50/day, CPL target with explicit scale trigger (2 weeks under target + 1 pilot call) and kill/fix trigger (CPL 1.5× target = creative problem, not budget problem).

Andromeda campaign (named for Meta's Oct-2025 ranking engine, whose #1 lever is creative diversity — >60%-similar creatives get suppressed):

• 1 campaign, Advantage+ on, CBO · 6 ad sets = customer-journey stages (Unaware → Problem-aware → Solution-aware → Product-aware/retargeting → Decision → Advocacy) · 5 ads per ad set = E-E-A-T angles: Experience ×2 (founder-voice + customer-scenario), Expertise, Authoritativeness, Trustworthiness.
• 30 distinct concepts, zero variants. Fatigued ad → new concept, not a recolor.
• Each ad brief: angle, hook line, format, CTA, destination. No creative production in this phase.
• Phased activation: cold stages first (they build the retargeting pool), retargeting at pool ≥1k, decision stage at scale trigger, advocacy only when real testimonials exist (never fabricate — leave the slot empty).
• Prerequisites checklist: capture page live, Pixel + CAPI deduplicated Lead/Schedule events, domain verified, seeded social pages. Check Special Ad Category applicability honestly (B2B software usually isn't; regulated-industry vocabulary may still trip review — keep creative explicitly B2B).

Gate: both docs exist; budget has scale AND kill triggers; every ad brief traces to a whitespace angle or E-E-A-T slot.

Phase 5 — 30-day content calendar + seed graphics

• Arc: Days 1–10 Whisper (problem + build-in-public, zero pitch) → 11–20 Tease (asset reveal, capture opens, limiting-beliefs series) → 21–30 Shout (pilot cohort, founding rate, countdown, referral). Give:ask ≥ 4:1 across the month.
• Voice: "How I," never "How to." One post/day written once, adapted per platform. Community/group posts are pure value, no links unless asked.
• Repurposing rule: organic winners (≥3× median saves/shares) become the next paid ad concepts.
• Seed graphics (adapt references/pipelines/build-graphics.mjs): ~10 cards, 1080×1350, brand tokens, HTML → chrome --headless=new --screenshot. Standard set: manifesto/wedge card, enforcement stat, regulatory timeline, checklist, two-column comparison, concept illustration (hand-drawn SVG beats stock), self-audit chain, guarantee card, whitepaper cover, countdown card. Read the rendered PNGs and fix spacing collisions before delivering — footer/caption crowding is the recurring bug.

Gate: calendar table complete (every row: phase, hook→substance, format+asset, CTA); all graphics rendered AND visually verified.

Phase 6 — Completion

• Deliverables table with complete absolute paths for every artifact (Cardinal Rule 8 — hard gate).
• Create tasks for the build steps that block launch (capture page, pixel, sending domain).
• Update project memory with what shipped and what's pending.
• If the whitepaper makes roadmap claims, ensure the roadmap build is a tracked task — the paper's credibility depends on the engine becoming real.

Known failure modes (learned the expensive way)

• Meta Ad Library: automated access is dead (403). Real logged-in Chrome via claude-in-chrome works. Keyword search ≠ advertiser search — always click the Page in the typeahead.
• Font files by hash: never assume a .woff2's identity from its filename — verify family/subfamily in the capture's fonts-manifest.json before embedding.
• Novelty claims: unhedged "first ever" is a falsification magnet. Always "to our knowledge" + name the nearest adjacent work.
• Honesty drift: marketing pressure blurs shipped vs. roadmap. The Phase 1 honest-claims list is binding; re-check every asset against it before delivering.
• Agent-tool prompt strings don't expand $(cat ...) — inline the content or have agents Read files themselves.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Stevekaplanai
• Source: Stevekaplanai/premium-product-launch-prompt
• License: MIT

Install and usage instructions live in the source repository linked above.