| name | ig-lead-simple |
| description | IG Lead Magnet Simple — v1 simplified loop. Asset creation → publish → comment-keyword trigger → DM Google Drive URL directly. Differs from v2: no Intercom, no email capture, no email sequence, no nurture; DM hands the link straight to the user. Carousel images use VPick `gpt-image-2`. Designed for "fast single-shot run, single-product PRD-driven" scenarios. Triggers on phrases like "ig-lead-simple", "simple IG lead magnet", "no email", "DM the drive link", "fast one-shot run", "follow my product PRD".
|
| metadata | {"methodology":"Alex Hormozi — $100M Leads, Lead Magnets + Content (simple track)","integrations":["boring","vpick"],"sister-skills":["ig-leadmagnet-machine","lead-magnet-pdf-creator","copywriting-bible/*"],"required_plugins":["100m-leads","copywriting-bible","marketing-agent"],"required_mcp":[{"name":"boring","usage":"IG publish + comment auto-reply + DM","note":"Self-hosted Boring instance. URL goes in your local config."},{"name":"vpick","tools":["run_image_generator"],"usage":"Carousel image generation (model: gpt-image-2)","note":"Self-hosted VPick instance. URL goes in your local config."},{"name":"obsidian-mcp","tools":["list_files","read_file","create_file","update_file","search_files"],"usage":"Read past campaign records + write this run's record (cross-machine portable, no local path dependency)","note":"Use any Obsidian MCP server (e.g. obsidian-rest, mcp-obsidian). See README."}],"not_required":["intercom","nano-banana"]} |
IG Lead Magnet Simple (v1)
Public release notice: this skill was extracted from a personal toolkit. All account IDs, MCP URLs, GCS bucket names, vault paths, and product names have been replaced with <PLACEHOLDER> tokens. See README.md for setup; see config.example.md for the full placeholder list.
The fastest IG lead-magnet loop: comment keyword → public reply → DM hands a Google Drive link directly. No email capture, no nurture sequence. You trade the CRM list for the lowest-friction validation speed.
Usage
/100m-leads:ig-simple <product> [language]
| Arg | Values | Default |
|---|
product | One of the products listed in your local config (see config.example.md). | required |
language | zh / zh-Hant (Traditional Chinese) or en (English) | zh |
🔴 Language hard rule: zh = Traditional Chinese (zh-Hant), simplified Chinese is forbidden. Every prompt, PDF body, caption, DM, and carousel text overlay must explicitly say Traditional Chinese / 繁體中文 so the LLM / gpt-image-2 doesn't render simplified.
Examples (the actual handles come from your config.example.md):
/100m-leads:ig-simple <your-product-A> zh → Traditional-Chinese run → IG <your_zh_handle> + TikTok <your_zh_tiktok_name>/100m-leads:ig-simple <your-product-B> en → English run → IG <your_en_handle> + TikTok <your_en_tiktok_name>
The skill will automatically:
- Read the product's PRD /
lead-magnet.md / design-guideline.md / campaign/*.md / image lists
- Apply the product-specific design system (color preset comes from
design-guideline.md, not the YAML default)
- Route to the language-matched IG + TikTok account (dual-platform synchronized publish)
- Run all 5 phases; pause for final confirmation right before Phase 4 publishes
The point of skipping interactive questions: stop re-asking "which product / which palette / which account" every run. Decisions are 90% automatic; you only pick the lead-magnet topic and hit publish.
┌─────────────────────────────────────────┐
│ parallel execution │
│ │
│ ┌──────────────┐ ┌──────────────┐ │
│ │ Phase 1 │ │ Phase 2 │ │
│ │ Lead Magnet │ │ IG Carousel │ │
│ │ PDF + Drive │ │ (gpt-image-2)│ │
│ └──────────────┘ └──────────────┘ │
│ │
│ ┌──────────────┐ │
│ │ Phase 3 │ │
│ │ Caption + DM │ │
│ └──────────────┘ │
└────────────────────┬────────────────────┘
▼
┌──────────────┐ ┌───────────┐
│ Phase 4 │ │ Phase 5 │
│ Boring publ. │ → │ Metrics │
│ +autoReply+DM│ │ │
└──────────────┘ └───────────┘
When to use this vs v2 (ig-leadmagnet-machine): you want to validate a hook / lead-magnet's market fit fast, you don't want to build an email nurture, the product is too early to justify a list. If you need a nurture sequence, switch to v2.
References
Phase 0: Pre-flight + product confirmation
🔴 Step 0.1: MCP connection health check
claude mcp list | grep -E "boring|vpick|obsidian"
All three must be ✓ Connected:
| MCP | URL | Purpose |
|---|
boring | <your-boring-host>/mcp/t/<token> | IG publish + auto-reply + DM |
vpick | <your-vpick-host>/mcp/t/<token> | Carousel images (gpt-image-2 @ 1K @ 4:5) |
obsidian-mcp | <your-obsidian-mcp-host>/mcp | Read/write Obsidian campaign records (cross-machine) |
If any shows ✗ Failed to connect:
claude mcp remove <name>
claude mcp add --transport http <name> <url>
First time on a new machine: run all three claude mcp add commands, then /exit and restart.
🔴 Step 0.1.5: Sandbox permission pre-check (required)
Auto-mode harness blocks "Create Public Surface" by default (publishing to IG, uploading public Drive files, calling public APIs). Phase 1 + Phase 4 hit these. Before running the skill, confirm .claude/settings.local.json has these 7 allow rules; surface what's missing for the user to copy-paste, then proceed to Phase 1.
Required allow rules (under permissions.allow):
"Bash(curl *www.googleapis.com/drive/v3/files*)",
"Bash(curl *upload/drive/v3/files*)",
"Bash(curl *tinyurl.com/api-create.php*)",
"Bash(gcloud auth print-access-token *)",
"mcp__boring__boring_publish_post",
"mcp__boring__boring_create_auto_reply_rule",
"mcp__boring__boring_update_auto_reply_rule"
Logic:
Read .claude/settings.local.json's permissions.allow array
→ List missing rules to user
→ User edits manually and saves (agent CANNOT self-add — self-modification is always blocked)
→ User says "ok" → proceed to Phase 0.2
⚠️ Don't try update-config skill to auto-add — same self-modification guard blocks it. Only the user can edit manually.
Step 0.2: Confirm product + language
Take <product> and [language=zh] from args. Only ask interactively if args are missing.
PRDs are read via the Obsidian MCP (path is relative to your vault root):
obsidian-mcp:read_file(path="<your-products-folder>/<product>/PRD.md")
Available products are defined in your local config.example.md — your products live under <your-products-folder>/<product>/PRD.md in your Obsidian vault.
If the user names a product not in your list, run obsidian-mcp:list_files(path="<your-products-folder>") to confirm; if no PRD found, ask back.
🔴 Step 0.2.4: Account routing (auto-select IG + TikTok by language)
Don't ask "which IG account" — it's deterministic from the language arg:
| Language | IG handle | IG account_id | TikTok name | TikTok account_id |
|---|
zh / zh-Hant (Traditional Chinese) | <your_zh_handle> | <IG_ZH_ACCOUNT_ID> | <your_zh_tiktok_name> | <TIKTOK_ZH_ACCOUNT_ID> |
en (English) | <your_en_handle> | <IG_EN_ACCOUNT_ID> | <your_en_tiktok_name> | <TIKTOK_EN_ACCOUNT_ID> |
The actual account IDs are in your local config.example.md (gitignored). Get them via boring_list_accounts and paste into your local copy.
🔴 zh = Traditional Chinese. Every instruction to LLM / gpt-image-2 must explicitly say "Traditional Chinese (繁體中文)" — never just "Chinese" (interpreted as simplified) and never "Simplified Chinese".
Dual-platform synchronized publish:
- IG: Carousel + auto-reply + DM Quick Reply (full loop)
- TikTok: same 7 images as photo slideshow broadcast (no auto-reply — Boring only supports fb/ig keyword rules). Caption adds a redirect line: "Want the PDF? Comment '' on my IG @" routing TikTok viewers to IG.
Write into BRIEF: account_handle uses the language-matched IG handle (for caption top-left).
Step 0.2.5: Read this product's past campaign records
Required step. Try obsidian-mcp MCP first, fallback to local Obsidian vault on read failure (your MCP server's vault may not be the same as your local vault):
# First try — obsidian-mcp (cross-machine friendly)
obsidian-mcp:list_files(path="<your-products-folder>/<product>/campaign")
obsidian-mcp:read_file(path="<your-products-folder>/<product>/campaign/<file>.md")
# On failure (ENOENT / empty array) → fallback to local Read
# Local vault path (note vault root often has an extra same-named folder):
Read("<YOUR_LOCAL_VAULT_ROOT>/<your-products-folder>/<product>/campaign/<file>.md")
⚠️ The Obsidian MCP server may run on a different machine from your local Obsidian app. They are typically not symlinked. Local edits won't be visible to MCP until rsync'd. If MCP read fails, just fallback locally — don't panic.
Each .md is one past campaign log (Hypothesis → Variables → Publishing Results → Learnings → Results → suggestions for next run).
Why read them: absorb prior lessons + adjust this run. Common takeaways:
- Last run's high-reach / high-comment hook → extend it
- Last run's flop hook → don't repeat
- Last run's prompt patterns that timed out / mis-rendered → preempt
- Last run's PDF chapter that resonated most with leads who stayed → strengthen
If list_files returns not-found / empty array: this is the product's first run; no need to pre-create the folder (create_file auto-creates intermediate dirs).
Output: a 1-2 sentence summary of "what last time taught me" into the Step 0.4 proposal (so the user sees you read history).
Step 0.2.6: Read "product images" + "protagonist photos" asset lists (cloud URLs)
Required step. The carousel design system needs two visual asset groups, both already on a public-readable GCS bucket. Same fallback rule as Step 0.2.5 — MCP first, local Read on failure:
# Product UI screenshots (embeddable on Solution / Method pages)
obsidian-mcp:read_file(path="<your-products-folder>/<product>/img-Product.md")
# fallback: Read("<YOUR_LOCAL_VAULT_ROOT>/<your-products-folder>/<product>/img-Product.md")
# Protagonist photos (cover, cross-page hero shots)
obsidian-mcp:read_file(path="<your-protagonist-folder>/protagonist-image.md")
# fallback: Read("<YOUR_LOCAL_VAULT_ROOT>/<your-protagonist-folder>/protagonist-image.md")
# Note: filename may have spaces — list_files to confirm
Md format convention (hard rule, see references/media-creation.md §1): each image entry is two lines — ![[wikilink]] (Obsidian preview) + → https://... (the cloud URL the carousel actually uses). Read the md, take only the line starting with → and feed that URL to vpick:upload_image.
Bucket structure (your own — define in config.example.md):
- Protagonist photos →
gs://<YOUR_GCS_BUCKET>/<YOUR_PROTAGONIST_PATH>/
- Product screenshots →
gs://<YOUR_GCS_BUCKET>/products/<product>/
Output:
Selection rule (see references/media-creation.md §1):
- Cover (page 1) + cross-page (5+6) → pick landscape
- Solution / Method page (4) → can embed 1-2 product screenshots
- Across the 7-page run, protagonist photo appears 2-4 times (per BRIEF.protagonist.photos_to_use)
If an image in the md doesn't yet have → URL → upload it to GCS first (gsutil -m cp ... gs://<YOUR_GCS_BUCKET>/products/<product>/), update the md, then proceed to Phase 2.
If img-Product.md doesn't exist → run list_files to confirm, then ask the user to add screenshots first.
🔴 Step 0.2.7: Read this product's design-guideline.md (required)
This is the product-specific visual design system, higher priority than the carousel-design-system.yaml default presets. Phase 2's image generation must apply it.
# Same fallback rule as 0.2.5: MCP first, local Read on failure
obsidian-mcp:read_file(path="<your-products-folder>/<product>/design-guideline.md")
# fallback: Read("<YOUR_LOCAL_VAULT_ROOT>/<your-products-folder>/<product>/design-guideline.md")
Extract from design-guideline.md (write into BRIEF):
| Section | Extract | Used in |
|---|
| Primary / Secondary / Accent colors | hex codes + roles | YAML PART 2 → color_scheme: custom slots |
| Brand mood / 1-line visual summary | e.g. "film studio + AI node canvas" | imageGenerator prompt's mood segment |
| Typography system | CJK + Latin font preferences | YAML PART 3 type tweaks |
| Typography rules | title size / leading / contrast | imageGenerator prompt's text overlay segment |
| Do / Don't visuals | e.g. "no toy aesthetic", "needs node-system feel" | imageGenerator prompt's style locking segment |
| Brand keywords | e.g. "Cinematic / Modular / Intelligent" | Caption / DM mood words |
If design-guideline.md doesn't exist → product hasn't built one. Ask the user to either run another skill to draft one, or fall back to the carousel-design-system.yaml default preset.
⚠️ Past lesson: a previous run completely skipped design-guideline.md and used the YAML default preset (ink_lemon — black + lemon yellow), but the product's actual design system was Electric Violet #7C3AED + Canvas Navy #111827 + Cinematic Cyan #22D3EE. Wrong palette. Always read the guideline.
Step 0.3: Read PRD, extract key info
After reading <product>/PRD.md, organize:
| Field | Source |
|---|
| Core pain points | ## 🔧 Problem section |
| Target audience | inferred from Problem + Why it matters |
| Value proposition | ## ✅ Solves section |
| Competitors + diff | ## 📊 Why it matters section |
| Product features | PRD middle (Solution / How it works) |
| Pricing | Why it matters table |
Step 0.3.5: Read Lead Magnet strategy page
Required step. Each product has a lead-magnet strategy doc listing 5 personas, 3 lead-magnet candidates per persona, prioritization, CTA design:
obsidian-mcp:read_file(path="<your-products-folder>/<product>/lead-magnet.md")
Organize for Step 0.4 proposal:
| Field | Source |
|---|
| Suitable lead magnet form | strategy doc opening "TL;DR" (Checklist / Template / Cheat Sheet / Blueprint / Demo) |
| Product core messages | "Product core messages" section (4-5 anchors to align lead magnets to) |
| 5 personas | each persona's "biggest pain" |
| 15 lead-magnet candidates | per persona × 3 |
| Top-3 priority | "Prioritization" section's first tier |
| CTA design | "Suggested CTA design" section |
Selection logic:
- Default: pick one from Top-3 priority (covers core audience, easy to extend into carousel + PDF)
- If user explicitly names a persona / topic → pick from that persona's 3 candidates
- Align with past campaigns (Step 0.2.5): which hook flopped, which drew comments → influences which to pick
If lead-magnet.md doesn't exist → product hasn't built a strategy. Ask back: run lead-magnet-creator skill first, or improvise from PRD pain points this once?
Step 0.4: Plan proposal (user confirmation)
Combine PRD extract + lead-magnet strategy into a concrete proposal, wait for user nod before Phase 1:
## IG Lead Magnet plan — [product]
**Lead magnet candidates** (from lead-magnet.md Top-3, user picks one or specifies):
1. "[First-tier title 1]" — form: [Checklist / Template / Cheat Sheet / ...]
Persona: [...]
Promise: [one line]
2. "[First-tier title 2]" — form: [...]
Persona: [...]
Promise: [...]
3. "[First-tier title 3]" — form: [...]
Persona: [...]
Promise: [...]
→ I recommend [N] because [reason from past campaigns / PRD].
**Selected lead magnet PDF**: [pick 1 of 3, or user-custom]
- Audience (Persona): [from lead-magnet.md]
- Problem solved: [PRD Problem #X + lead-magnet.md pain]
- Planned chapters: [expand the chosen candidate's "content" into 3-5 sections]
- CTA: [pick 1-2 from lead-magnet.md "Suggested CTA design"]
**Publishing channels** (auto-routed by language — Step 0.2.4):
- IG: <your_zh_handle / your_en_handle>
- TikTok: <your_zh_tiktok_name / your_en_tiktok_name> (cross-post photo slideshow, caption redirects to IG comments)
**IG carousel plan** (immersive 7-page, applying `design-guideline.md` > `carousel-design-system.yaml`):
- 7 pages (Cover / Hook Reframe / Stakes / Method 1 / Method 2 / Method 3 / CTA)
- pages 5+6 are a cross-page (one image cut into two halves)
- Palette: **extract from design-guideline.md** (product-specific) → write into YAML PART 2 `custom`
- Fall back to one of 4 yaml presets only if no design-guideline.md
- Protagonist photos: ×[2-4] from `<your-protagonist-folder>/protagonist-image.md` → upload as VPick reference in Phase 2.0
- Product UI screenshots: [1-2 from `<your-products-folder>/<product>/img-Product.md`] → upload as VPick reference in Phase 2.0
- Cover hook: "[pick from pain / desire / mystery]"
- Trigger keyword: "[short memorable, ≤ 4 zh chars / ≤ 8 en chars]"
**BRIEF summary** (YAML PART 1):
- series_name: [series — reuse from last campaign or new]
- account_handle: @<language-matched IG handle>
- product_name: [...]
- topic: [...]
- color_scheme: `custom` (filled with design-guideline.md hex)
- protagonist.photos_to_use: 2-4
- cta.keyword / deliverable: [keyword / what we're sending]
**DM payload**:
- Copy intro + Google Drive PDF link
- Expected CTA: tap link → reads PDF immediately
Confirm direction OK and I'll run Phase 1/2/3 in parallel.
Phase 1: Lead Magnet PDF + Google Drive upload
🔴 Language hard rule:
language=zh → PDF entirely in Traditional Chinese (zh-Hant), HTML uses <html lang="zh-Hant">, fonts include Noto Serif TC + Noto Sans TC. No simplified.language=en → entirely English, HTML <html lang="en">, fonts Inter / Playfair Display- Every LLM prompt for copy must say
output in Traditional Chinese (zh-Hant) or output in English, never just "Chinese"
Step 1.1: Design PDF content (using lead-magnet.md selected option)
Invoke lead-magnet-pdf-creator skill (from the marketing-agent plugin):
- Source: Step 0.3.5 selected option (title / form / promise / outline / CTA)
- Reinforce with: PRD pain points, solutions, features
- Built-in Sugarman copywriting engine auto-optimizes
- Outputs a brand-styled PDF
Content structure (apply lead-magnet.md's chosen option's "content" as chapters; tweak per form):
| Form | PDF structure |
|---|
| Checklist (most common) | Cover → ch 1-N each one checklist item → final CTA |
| Template / Prompt Pack | Cover → templates 1-N (with use case + sample output) → final CTA |
| Cheat Sheet | Cover → one comparison table + usage notes → final CTA |
| Blueprint / Workflow | Cover → flowchart → each phase breakdown → final CTA |
| Walkthrough | Cover → step-by-step (with screenshots / code) → final CTA |
Generic skeleton (tweak per form):
Cover — title + promise subtitle (from lead-magnet.md option)
↓
[chapters per form, content from lead-magnet.md "content" section]
↓
Final CTA — pick 1-2 from lead-magnet.md "Suggested CTA design"
(CTA 1 signup / CTA 2 try directly / CTA 3 nurture)
Don't always use a fixed "Problem → Failure → Solution → Steps" 5-chapter mold. Form follows lead-magnet.md — Checklist is pure Checklist, Cheat Sheet is one comparison table.
Step 1.2: Upload to Google Drive, get public link
Full API flow + helper script: references/google-drive-upload.md
Architecture: each lead magnet creates a <product>-YYYYMMDD subfolder under your parent folder (<YOUR_DRIVE_PARENT_FOLDER_ID>) → set anyone-with-link → drop PDF in → use https://drive.google.com/file/d/<FILE_ID>/view for the DM.
Three core steps (curl + Drive API; token from gcloud auth print-access-token --account=<your-gcloud-account>):
TOKEN=$(gcloud auth print-access-token --account=<your-gcloud-account>)
PARENT="<YOUR_DRIVE_PARENT_FOLDER_ID>"
FOLDER_ID=$(curl -s -X POST "https://www.googleapis.com/drive/v3/files" \
-H "Authorization: Bearer $TOKEN" -H "Content-Type: application/json" \
-d "{\"name\":\"<product>-$(date +%Y%m%d)\",\"mimeType\":\"application/vnd.google-apps.folder\",\"parents\":[\"$PARENT\"]}" \
| python3 -c "import sys,json;print(json.load(sys.stdin)['id'])")
curl -s -X POST "https://www.googleapis.com/drive/v3/files/$FOLDER_ID/permissions" \
-H "Authorization: Bearer $TOKEN" -H "Content-Type: application/json" \
-d '{"role":"reader","type":"anyone"}'
FILE_ID=$(curl -s -X POST "https://www.googleapis.com/upload/drive/v3/files?uploadType=multipart" \
-H "Authorization: Bearer $TOKEN" \
-F "metadata={\"name\":\"lead-magnet.pdf\",\"parents\":[\"$FOLDER_ID\"]};type=application/json" \
-F "file=@/tmp/lead-magnet.pdf;type=application/pdf" \
| python3 -c "import sys,json;print(json.load(sys.stdin)['id'])")
echo "DM URL: https://drive.google.com/file/d/$FILE_ID/view"
⚠️ Long URL displays poorly in IG DM → shorten with bit.ly / TinyURL before Phase 3 (see reference).
⚠️ First-time setup: if token returns 403 ACCESS_TOKEN_SCOPE_INSUFFICIENT, run gcloud auth login --enable-gdrive-access once.
Output: https://drive.google.com/file/d/<FILE_ID>/view (or shortened) → used in Phase 3.
Phase 2: IG Carousel (VPick gpt-image-2, immersive 7 pages)
Authoritative design source: references/carousel-design-system.yaml
Phase 2 follows this YAML. Below is just the flow summary — details (PART 3 type / PART 4 grid / PART 5 visuals / PART 7 7-page narrative) live in the YAML.
🔴 Language hard rule (every imageGenerator prompt):
zh → Display Traditional Chinese text (繁體中文 / zh-Hant): "..." — never just "Chinese text" (gpt-image-2 will render simplified)en → Display English text: "..."- Fonts: zh uses
Noto Serif TC Bold (titles) / Noto Sans TC Medium (body) / Inter Medium (English labels); en uses Playfair Display or Inter Bold (titles) / Inter (body)
🔴 Step 2.0: Upload reference images (required)
Before any imageGenerator runs, upload protagonist + product UI as VPick upload nodes. imageGenerator prompts then reference them via @protagonist_ref @product_ref + connect_nodes, producing the real protagonist face and real product UI instead of AI-imagined versions.
# 1. Create / switch into a VPick project for this run
vpick:create_project(name="<product>-<topic>-<date>")
# 2. Upload protagonist photo (URL from Step 0.2.6 → URL)
vpick:upload_image(
url="<protagonist photo cloud URL from your protagonist-image.md>",
name="protagonist_ref"
)
# 3. Upload product UI screenshot (most representative 1-2)
vpick:upload_image(
url="<product screenshot cloud URL from your img-Product.md>",
name="product_ref"
)
Which pages need references:
| Slide | Reference | Why |
|---|
| 1 Cover (image_full_bg) | @protagonist_ref | Cover needs the real face — random AI face has no recognition |
| 5 Method 2 cross-page L | @protagonist_ref | Sequential shot, must match cover's person |
| 4 Method 1 (with product UI) | @product_ref | Embed real UI, not AI imagination |
| 6 Method 3 cross-page R | @product_ref (optional) | Topic-dependent |
imageGenerator prompt pattern:
[scene description...]
Reference image @protagonist_ref for the protagonist's face and styling — keep facial features identical.
[text overlay segment...]
When add_node's prompt contains @protagonist_ref, VPick auto-connect_nodes to the upload node (Element system).
🟡 Step 2.0.5: VPick active-project guard
VPick's active project is user-level state — opening another tab in VPick can switch the active project. Verify before each op:
# Before add_node / run_image_generator / list_nodes:
vpick:list_projects → find isActive: true → compare against targetProjectId
# If mismatch:
vpick:switch_project(projectId=<targetProjectId>)
Most-common stumbling moment: mid-batch (4+3 in flight), user opens another VPick project in another tab → subsequent nodes land in the wrong project.
Step 2.1: Fill BRIEF into YAML PART 1
From Phase 0.4 proposal + PRD extract, fill the YAML brief (mental model is fine; don't write a file):
| YAML field | Source |
|---|
series_name | custom or reuse from last campaign |
account_handle | language-matched IG handle |
product_name | Phase 0.2 confirmed product |
topic | this run's topic (from PRD pain) |
protagonist.photo_style | e.g. "cinematic, low-saturation, study or studio setting" |
protagonist.photos_to_use | 2-4, default 3 |
audience.persona / pain_point / desired_outcome | from PRD |
color_scheme | from design-guideline.md custom (preferred) or one of 4 presets |
cta.keyword / deliverable | comment trigger + what's delivered |
Step 2.2: 7-page narrative breakdown (YAML PART 7)
1. Cover — Cover hook (pain / desire / mystery), full-bleed protagonist landscape
2. Hook Reframe — "you're not doing X, you're doing Y" reframe, secondary solid bg
3. Stakes — "{element} is the only {make-or-break / dividing line}", primary solid bg
4. Method 1 — "{action 1}, {action 2}, {action 3}" 3 short verbs, can embed product UI
5. Method 2 — cross-page LEFT (image_full_bg_left), technique + mechanism
6. Method 3 — cross-page RIGHT (image_full_bg_right), metaphor wrap
7. CTA — desire-language CTA + "comment '<keyword>' for <deliverable>", primary solid bg
Color cadence (YAML PART 6): page 1 image → 2 secondary → 3 primary → 4 secondary → 5+6 cross-page image → 7 primary.
Cover headline scoring: use the headline-lab skill, ≥ 20/25 to ship.
Product image embedding: page 4 / 5 can embed img-Product.md selections (contextual treatment, not naked screenshots).
Step 2.3: VPick image generation (4+3 parallel batches)
Use vpick:run_image_generator with gpt-image-2, fixed 1K + 4:5:
vpick:add_node:
type: "imageGenerator"
name: "Slide{N}_{Role}"
model: "gpt-image-2"
prompt: "[per media-creation.md §3 4-segment template + YAML PART 7 page template]"
data: { aspectRatio: "4:5", resolution: "1K" }
Hard rule: Carousel images use VPick gpt-image-2 @ 1K + 4:5. No fallback to nano-banana. If layout is bad, rewrite the prompt or hand-fix in a design tool — never switch models.
Prompt 4-segment (details in references/media-creation.md §3):
- Scene: per BRIEF.protagonist.photo_style + page role (image_full_bg / primary / secondary)
- Text overlay:
Display [language] text: "..." + position + font (Noto Serif TC Bold / Inter Medium) + key 2-4-word highlight color block (YAML PART 5)
- Layout: top-left series_name + handle, top-right page pill
{N}/7, bottom gradient mask (required on image_full_bg pages)
- Style lock: YAML palette mood + primary/secondary hex + ban emoji / exclamation / stock photo
Generation strategy:
- 7 images split into 4 + 3 parallel batches
- Cross-page 5+6 in same batch, prompts cross-reference "right edge cuts off" / "left edge continues from prior" + composition negative space
- Each completion auto-uploads, returns URL
Step 2.4: Order the 7 URLs
Arrange by slide order:
slide_urls = [
"https://...slide1-cover.png",
"https://...slide2-hook-reframe.png",
"https://...slide3-stakes.png",
"https://...slide4-method1.png",
"https://...slide5-method2-left.png",
"https://...slide6-method3-right.png",
"https://...slide7-cta.png",
]
If VPick's URL is rejected by IG, re-host via boring:boring_upload_from_url to get a Boring-accepted public URL.
Phase 3: Caption + keyword + DM copy
🔴 Language hard rule:
zh → caption / keyword / public reply / DM intro / full DM body all Traditional Chinese. Use TW/HK conventions (e.g. 「影片」 not 「视频」, 「軟體」 not 「软件」, 「使用者」 not 「用户」)en → all English, no zh/en mixing- When invoking copywriting-bible sub-skills, prompt must say:
produce all output in Traditional Chinese (zh-Hant) — Taiwan / HK conventions, NOT Simplified Chinese
🎯 3.0 CTA philosophy (read before writing copy)
Use desire language, not fear language. Pain can anchor attention, but only desire triggers action.
Three elements
- Familiar scene — anchor to their daily moment: morning coffee, scrolling IG, end-of-day collapse, weekend "I want to do something but not all day"
- Already-desired outcome — make their wanted result concrete and visible: video shipped, client commenting in group, friends asking how, finishing on time
- Easier than they think — preempt the "this sounds great but the bar must be high" reflex: "one sentence → done", "no new UI to learn", "tools you already use"
❌ Fear version (don't write this)
- "Keep using Runway and you'll keep wasting time"
- "Don't solve this and your videos will never ship"
- "Your competitors are already on AI; you're still hand-cutting timelines"
- "I know [pain] is really frustrating..." ← still fear language
✅ Desire version
Imagine tomorrow morning, you open your laptop, tell Claude "make me a 5-shot car ad" and walk to the kitchen for coffee.
By the time you sit back down — 6 storyboard images, 5 video clips, one audio track are arranged on the canvas. Hit publish.
No new software. No 7-tool tab-switching. No 2 AM cuts.
What you thought would take half a year of practice? Today, one sentence kicks it off.
⤷ This hits all three: ① familiar scene (morning, coffee) ② desired outcome (automated result) ③ easier than thought (one sentence, no learning).
3.1 Caption (apply slippery-slide-writer)
[Hook — first line ≤ 10 zh chars / ≤ 12 en words] (first-sentence-crafter)
↓ Hook can carry pain to anchor attention
[Pain amplification — 1-2 lines ≤ 30%]
short, only does the "yes I feel that too" resonance
[Bridge]
"But..." / "Until I found..." / "Then..."
[Desire scene — 2-3 lines ≥ 70% the real engine]
familiar scene + desired outcome + easier-than-thought
[Value preview — 3-5 bullets]
use "you'll", "you can", "you will" (capability framing)
[CTA — desire language] (emotional-sell-rational-justify)
👇 Comment "<keyword>" and I'll DM you the full guide
#hashtag1 #hashtag2 ... (5-10)
Ratio: pain ≤ 30%, desire + action ≥ 70%. Pain anchors; desire drives.
3.2 Keyword choice
Short, memorable, hard to misspell. Pick something with desire/action energy rather than vague generic verbs:
| Good | Better (desire-flavored) | Bad |
|---|
YES, SEND, FREE | EDIT, BUILD, WIN, KIT, GUIDE | long phrases, mixed-language |
3.3 Auto-reply + DM copy (core diff vs v2)
Public auto-reply (other commenters see "yes this really sends" → FOMO):
Sent to your DMs ✉️
Note: not "thanks for your interest" (sounds desperate). Use a completed-action sentence.
DM private (direct Drive link, all desire language):
[1. Friendly intro — not subservient]
Hey 👋 glad you grabbed this one.
[2. Desire scene — 1-2 lines, NOT "I know you're struggling"]
The thing I love most about it: [familiar scene + desired outcome].
I thought it would take [assumed high bar], but [the easier truth].
[3. Direct delivery — Google Drive URL]
Here's the full guide (free PDF, no signup):
👉 [Google Drive public link]
[4. Value preview — "you can / you'll / you will" framing]
Inside you'll find:
✓ You can [outcome from chapter 1]
✓ You'll learn [capability from chapter 2]
✓ You'll get [specific result] in [specific time] (chapter 3)
[5. Soft CTA — open invitation, not pressure]
Ping me anytime if you want to chat.
Or jump straight in: [product URL] — [one desire line].
Pre-send checklist:
vs v2: v2's DM is an Intercom landing page (must enter email); this version's DM is the Drive PDF directly — tap and read.
Phase 4: Boring publish + automation
🔴 Confirm 3 things upfront with the user: the harness splits each external write into separate approvals. A single "ok" from the user only lets through the first. Be explicit:
"Phase 4 has 3 external writes: ① publish IG carousel ② create IG auto-reply + DM rule ③ cross-post to TikTok. Reply 'send' to authorize all 3 in one go."
Step 4.1a: Publish carousel to IG (language-matched account)
boring:boring_publish_post:
account_id: "<language-matched IG account_id (Step 0.2.4)>"
platform: "instagram"
text: "[Phase 3.1 caption]"
media_urls: [Phase 2.4 7 URLs]
🔴 Step 4.1.5: Get post_url (required — used in 4.2)
After publish, IG returns media_id but no post_url. boring_create_auto_reply_rule server-side needs post_url to bind the rule. Right after publish, call:
boring:boring_get_publish_history(platform="instagram", limit=3)
# → find the entry matching submission_id, take post_url
# Format: https://www.instagram.com/p/<shortcode>/
🔴 Step 4.1b: Cross-post to TikTok (same 7 images as photo slideshow)
Don't regenerate. Use the same 7 carousel URLs on TikTok. But rewrite the caption — TikTok has no keyword auto-reply, so redirect viewers to IG comments to get the PDF.
TIKTOK_CAPTION = """
[1-2 line hook, same as IG]
⬇️ Want the full guide PDF?
Comment "<keyword>" on my IG @<language-matched IG handle>
and I'll DM you the whole thing 📩
[remaining caption...]
"""
boring:boring_publish_post:
account_id: "<language-matched TikTok account_id (Step 0.2.4)>"
platform: "tiktok"
text: TIKTOK_CAPTION
media_urls: [same 7 URLs as IG]
draft: false
⚠️ TikTok doesn't support Boring's keyword auto-reply (only fb / ig). TikTok is broadcast-only — caption funnels traffic to IG.
⚠️ If TikTok rejects an image URL (hash already used), re-host with boring:boring_upload_from_url.
Step 4.2: Set comment auto-reply + DM
Conflict check:
boring:boring_list_auto_reply_rules
# Same account + same keyword → boring_delete_auto_reply_rule then recreate
# Different accounts or different post_urls can coexist (rules bind to post_url)
Create rule (post_url is required, account_id matches language):
boring:boring_create_auto_reply_rule:
account_id: "<language-matched IG account_id (Step 0.2.4)>"
platform: "instagram"
post_url: "[IG post URL from Step 4.1.5]" ← required
keyword: "[Phase 3.2 keyword]"
reply_message: "[Phase 3.3 public reply]"
dm_message: "[Phase 3.3 DM intro, with Quick Reply]"
dm_quick_replies:
- title: "Send the PDF"
payload: "send_link"
dm_quick_reply_responses:
send_link: "[Phase 3.3 full DM with Drive PDF URL]"
⚠️ If the MCP tool schema doesn't list post_url but the server requires it → schema is stale, /exit and restart the session to pull a fresh schema.
Step 4.3: Test the loop
After publish, manually test:
- Use a second IG account, comment the keyword
- Public auto-reply ✓
- DM received with intro + Quick Reply button ✓
- Tap Quick Reply → full DM + Drive link ✓
- Tap Drive link → PDF opens ✓
If any step fails, fix it now — don't wait for real comments to discover it's broken.
Step 4.4: Write campaign log to Obsidian
Required step, every run. This is for the next run to read; skipping = wasted run.
Use obsidian-mcp MCP (don't use local Write — the skill is designed cross-machine and local paths aren't portable; MCP is the only guaranteed cross-machine path):
obsidian-mcp:create_file(
path="<your-products-folder>/<product>/campaign/<YYYY-MM-DD>-ig-lead-simple.md",
content="<frontmatter + 11 sections markdown>"
)
Path is relative to vault root (vault root is controlled by your obsidian-mcp server). create_file auto-creates intermediate dirs, but errors if the file exists — for same-day re-runs, use update_file or rename -v2, -v3.
Required frontmatter
---
date: YYYY-MM-DD
skill: ig-lead-simple
product: <product>
status: published
platform: instagram
ig_account: <your_zh_handle / your_en_handle>
post_id: <Boring's media_id>
keyword: <trigger keyword>
---
Required sections
| Section | What to write |
|---|
| Hypothesis | This run's core bet — which hook, which lead magnet, which keyword |
| Lead Magnet Choice | Which option from lead-magnet.md (title + form + persona) + why this vs the other Top-3 |
| Variables | All controllable variables (hook type / pages / model / resolution / keyword / DM flow / cross-post) |
| Content Plan | PDF chapters + Carousel 7-page plan (per yaml) + Caption structure |
| BRIEF Snapshot | YAML PART 1 values you used |
| Image Prompt Base | Common prompt patterns + design rules (so next run avoids the traps) |
| Expected Outcome | Reach, comment rate, DM trigger, conversion baselines |
| Status | published |
| Publishing Results | post_id, submission_id, auto-reply rule_id, publish time, account, Drive PDF URL, 7 carousel URLs |
| Learnings | What broke this run, the winning move, what next run should preempt |
| Results (fill in 72h later) | 6 empty checkboxes: reach, comments, DMs, Quick Reply taps, signups, paid |
| Next Action | D+1 / D+3 / D+7 / D+30 — what to check, what to fill back |
| Suggestions for next campaign | Explicit do/don't for future-you (most important section) |
Follow-ups
- D+3: run
boring_get_posts_performance + boring_list_auto_reply_logs, fill Results
- D+7: compare with same-product v2 records (if any) for a diff
- D+30: add Stripe signup / paid / retention numbers
Phase 5: Basic metrics
No email capture = no funnel mid/lower data. This version sees only the IG end.
Core metrics (Boring MCP)
boring:boring_get_posts_performance → post analytics
boring:boring_list_auto_reply_logs → keyword trigger log
| Metric | Baseline | Optimization |
|---|
| Reach | — | Hook + hashtags + posting time |
| Comment rate | ≥ 3% | Cover hook + lead magnet attractiveness |
| Read-through (slide 7/7) | IG insights | Cross-page 5+6 hooked them, CTA strong enough to stop |
| DM delivery rate | ≥ 95% | Auto-reply rule setup |
| Drive PDF click rate | Drive Activity | DM copy persuasiveness |
Want to know who downloaded: change Drive permissions to "must sign in to view" — but adds friction. Simple defaults to no friction; if you want this, upgrade to v2.
Signals to upgrade to v2
If you hit any of these, switch to ig-leadmagnet-machine (v2):
- Want to know who downloaded → email capture
- Want a nurture sequence → email automation
- Want to A/B test lead magnets → landing page splits
- Want CRM / GA4 integration → Intercom
Output checklist
| # | Item | Tool |
|---|
| 0 | Past campaign records review | obsidian-mcp list_files + read_file |
| 1a | Product PRD extract | obsidian-mcp read <your-products-folder>/<product>/PRD.md |
| 1b | Lead magnet selected (Top-3 → 1) | obsidian-mcp read <your-products-folder>/<product>/lead-magnet.md |
| 2 | Lead magnet PDF | lead-magnet-pdf-creator skill |
| 3 | Google Drive public link | Drive API + URL shortener |
| 4 | 7 carousel images (4:5 1K, immersive) | VPick run_image_generator (gpt-image-2) + carousel-design-system.yaml |
| 5 | Caption + DM copy | copywriting-bible/* skills |
| 6 | IG carousel published | Boring boring_publish_post |
| 7 | Auto-reply + DM Quick Reply | Boring boring_create_auto_reply_rule |
| 8 | Loop tested | manual: comment → DM → tap Drive → see PDF |
| 9 | Campaign log written to Obsidian | obsidian-mcp create_file <product>/campaign/<date>-ig-lead-simple.md |
Core principles (Simple ethos)
- Speed > completeness: validate hook/PDF first, invest in email later
- Zero friction: one tap on Drive link → PDF; no email, no signup
- PRD-driven content: every run starts from one product's PRD; PDF + caption serve that product's conversion
- Run fast, throw fast: failed hooks get cut, no sentimentality
Troubleshooting
MCP connection (most common)
| Symptom | Fix |
|---|
| ToolSearch can't find vpick / boring tools | claude mcp add then /exit to restart session |
claude mcp list shows ✗ Failed | claude mcp remove → add → /exit |
| MCP disappears mid-conversation | Immediate /exit and restart, don't retry in same session |
VPick gpt-image-2 issues
| Symptom | Fix |
|---|
| Text typos / mis-rendering | Rewrite prompt: Display [language] text: "...", avoid long sentences |
| Some of 7 parallel runs fail | Switch to 4+3 batch, retry failed ones individually |
| Style drift | Always lead prompt with YAML palette mood + primary/secondary hex |
| Cross-page 5+6 misaligned | Same-batch run, prompts cross-reference "right edge cuts" / "left edge continues" + composition; if still off, hand-fix in design tool |
| gpt-image-2 typography looks bad | Text-heavy pages (CTA): rewrite prompt (short lines + explicit position), or hand-fix in design tool. No fallback to nano-banana |
Drive / upload
| Symptom | Fix |
|---|
| Drive link asks for login | Re-check Step 2 permissions (type:anyone, role:reader) |
| URL too long for IG DM | bit.ly / TinyURL |
| No gdrive CLI | Use GCS instead (gsutil cp) — https://storage.googleapis.com/... is also public |
Boring auto-reply
| Symptom | Fix |
|---|
boring_create_auto_reply_rule returns 400 "keyword exists" | list_auto_reply_rules confirm → delete → recreate |
| DM received but no Quick Reply button | Check dm_quick_replies array format, title ≤ 13 chars |
| Comment triggered but no DM | Check IG account is business + DM permission granted |
