Lead magnet factory

Lead Magnet Factory

Commercial pipeline that turns a niche + topic into a publishable 9-page listicle ebook, matching the Marie Forleo "12 Mistakes" format. Output is a Canva Bulk Create CSV + human-readable Markdown preview — user pastes CSV into their Canva template, reviews, exports PDF.

When to invoke

User intent examples (trigger keywords in Vietnamese + English):

• "làm lead magnet" / "tạo ebook" / "làm ebook 12 điều"
• "make a lead magnet" / "create a listicle ebook"
• "build a PDF freebie for my course"
• Any request for a numbered-tips PDF ebook (9–12 items)

Do NOT use this skill for: long-form ebooks (>20 pages), fiction, technical whitepapers, academic content. Those need different pipelines.

Required inputs (ask user if missing)

1. Niche — e.g., "personal branding coaching cho creator Việt"
2. Topic/angle — e.g., "12 sai lầm khi build IG personal brand"
3. Target audience — persona: demographic + pain point + aspiration
4. Product CTA — what course/coaching/product the ebook sells into
5. Language — vi or en
6. Brand voice notes (optional) — if missing, default to Marie Forleo style in reference/style_guide.md
7. Author name + signature style — for final page
8. Photo asset inventory (optional) — list of photo filenames user has (e.g., portrait_laptop.jpg, standing_smile.jpg) so visual brief maps to real assets

If user provides fewer than 4 inputs, ask ONE consolidated question to fill all gaps. Do not go back-and-forth.

7-phase workflow

Execute phases sequentially. Use TodoWrite to track progress visibly to user.

Phase 1 — Market Research (Tavily MCP)

Goal: Surface 25–40 candidate pain points / mistakes / "stop doing X" items.

Actions:

1. Call mcp__tavily__tavily_research with queries in the target language:
   • "{niche} common mistakes 2026"
   • "{target_audience} biggest struggles {topic}"
   • "{niche} complaints reddit quora"
   • For Vietnamese: also search "{niche} Vietnam" to surface local context
2. If user already has Airtable base appZgsckgOCllyYjW with competitor intel, pull viral content + pain point data via Airtable MCP (list_records on Pain_Points or Content table)
3. Extract candidate items into output/{slug}/research_raw.json:

   [{"candidate": "text", "source": "url", "frequency_signal": "high|med|low", "emotion": "fear|frustration|aspiration"}]
   

Quality gate: ≥ 20 candidates before proceeding. If fewer, run 2 more Tavily queries.

Phase 2 — Angle Selection (Claude reasoning)

Goal: Pick exactly 12 items that form a coherent narrative arc.

Scoring rubric (apply to each candidate):

• Pain intensity (1–5): how much does the reader suffer from this?
• Specificity (1–5): is it a concrete "stop X" or vague?
• CTA fit (1–5): does solving this naturally lead to the product CTA?
• Variety (1–5): does it add new angle vs already-selected items?

Select top 12 by composite score. Ensure narrative flow: open with mindset items → middle with tactical items → close with transformation/bigger-picture items (match Marie Forleo arc).

Write selection + rationale to output/{slug}/selection.md for review traceability.

Phase 3 — Outline & Schema

For each of 12 items, generate outline:

• heading: "Stop {verb} {object}." — imperative, < 8 words
• hook_sentence: one punchy line (the "reframe")
• body_outline: 2 paragraphs — paragraph 1 = expose the problem/myth, paragraph 2 = actionable reframe + CTA internal-link

Also outline:

• Cover title + subtitle + highlight words
• Intro hook (2 paragraphs)
• Offer page (title, description, 3 benefits)
• Testimonials placeholder (5 slots — user fills real testimonials later)
• Closing page (4 differentiators + final CTA + signature)

Phase 4 — Drafting (Claude Sonnet 4.6)

Write full content following reference/style_guide.md:

• Imperative, punchy sentences
• "You"-focused address
• Mix short + medium sentences (avoid walls of text)
• Use emoji markers sparingly (👉 💖 💼 🌍 💡) per Marie Forleo pattern
• Each item body: 40–80 words total across 2 paragraphs (Canva cell constraint)
• Signature power phrases where natural (avoid cliché: no more than 1 "the truth is" per ebook)

Output conforms to schema.json. Write to output/{slug}/content.json.

Phase 5 — Editorial Pass (Opus 4.7 1M context if available, else Sonnet 4.6)

Read entire JSON in one context. Check:

• No item heading repeats a verb/object combo from another item
• Voice consistent across all 12 items
• No hallucinated statistics or fake testimonials (testimonial slots must be placeholders like [testimonial_1_placeholder])
• CTA threads through: intro hints at it, items 6/12 foreshadow, offer page delivers
• Word counts within bounds (see scripts/validate.py)
• Language consistency (all Vietnamese or all English, no code-switching unless intentional)

Fix issues in-place. Re-write content.json if edits made.

Phase 6 — Visual Brief (image mapping)

For each of 9 pages, produce visual instruction:

• If user provided photo_asset_inventory: map each page to best-fit filename (e.g., page 2 hook → portrait_laptop.jpg)
• If not: generate AI image prompt for Ideogram v3 (infographic/text-in-image) or Flux 2 (photorealistic)
• For cover + closing: always portrait of author (user's own photo preferred)

Write to output/{slug}/visual_brief.md with format:

## Page 1 — Cover
Asset: portrait_confident.jpg (or AI prompt if no asset)
Treatment: lime green stroke around figure, purple background card
Notes: title overlays with orange pill-highlight on key word

Phase 7 — Export

Run scripts in order:

1. python scripts/validate.py output/{slug}/content.json — must pass all gates
2. python scripts/to_canva_csv.py output/{slug}/content.json output/{slug}/canva_bulk.csv
3. Generate human-readable output/{slug}/preview.md by formatting content.json into Markdown
4. If Airtable MCP available, upsert summary row to Ebooks table (create table if missing): {slug, niche, topic, generated_at, status: "draft_ready"}

Final message to user must include:

• Path to CSV for Canva Bulk Create
• Path to Markdown preview for review
• Path to visual brief
• Next-step instructions (3 steps max): open Canva template → Bulk Create → upload CSV

Directory layout (per ebook run)

output/
  {slug}/                          # e.g., 12-sai-lam-ig-2026
    research_raw.json              # Phase 1
    selection.md                   # Phase 2
    content.json                   # Phase 3–5 (final structured)
    visual_brief.md                # Phase 6
    canva_bulk.csv                 # Phase 7 — UPLOAD THIS TO CANVA
    preview.md                     # Phase 7 — REVIEW THIS

Commercial reliability rules

1. Never fabricate testimonials. Use placeholders [testimonial_1_placeholder] — user replaces with real ones before publishing.
2. Never hallucinate statistics. If Phase 1 research didn't surface a stat, don't invent one. Use qualitative language instead.
3. Cite sources in selection.md for every item that came from a specific research source.
4. Language lock. Do not code-switch between Vietnamese and English inside body text unless brand voice explicitly allows it.
5. Copyright clean. Headlines and phrasing must be original — do not copy Marie Forleo's exact wording; use her structural template only.
6. Legal/health/finance disclaimer. If niche touches health, finance, or legal — append a disclaimer line to the closing page.

Cost budget per run

Expected token usage (Sonnet 4.6 for drafting, Opus 4.7 for final pass):

• Research: ~5K in, ~3K out
• Selection + Outline: ~3K in, ~4K out
• Drafting: ~5K in, ~3K out
• Editorial: ~10K in (full doc), ~3K out
• Visual brief: ~3K in, ~2K out

Total: ~26K input + 15K output ≈ $0.50–$1.00/ebook. Flag user if a run exceeds $2.

Failure modes to watch

• Tavily returns thin results → widen queries, don't fabricate. If still thin, ask user for seed pain points.
• Items 1–12 feel repetitive → in Phase 5, detect and rewrite. Never ship with duplicate angles.
• CTA doesn't thread → rewrite offer page to match final item's handoff line.
• Language/voice drift → Phase 5 must catch this. If drift is systemic, re-run Phase 4 with stricter style guide citation.

Reference files (read at invocation)

• reference/style_guide.md — voice, sentence patterns, emoji conventions
• reference/canva_template_spec.md — exact placeholder names expected by user's Canva template
• schema.json — strict output schema
• examples/sample_output.json — reference Vietnamese example

Load reference/style_guide.md into context at Phase 4 start. Load schema.json at Phase 3 start.