The Shelf / Growth / Short-Form Video Growth Engine
Short-Form Video Growth Engine
Faceless short-form for app installs: pillars, pain-tied hooks, scripts, a calendar.
The job: turn short-form video — TikTok, Reels, and Shorts — into app installs, without ever standing in front of a camera. The skill turns your value props, ICP, and top user pains into a whole content system: pillars, a hook bank, shot-listed faceless scripts, captions, a CTA→install path, and a dated calendar. It plans; you (or an editor) shoot.
Most short-form advice fails one of three ways. You post at random and blame the algorithm for the silence. You paste a generic "give me viral content" prompt and get hooks that could sell anything, which means they sell nothing. Or you buy a "faceless mastery" course that teaches format names and leaves you to write the actual scripts. All three skip the only part that matters: the hook has to be tied to your app's specific value and your user's specific pain, or it's noise in a crowded feed.
What's on the tag:
- A value⇄pain matrix, then 3–4 content pillars — each assigned a faceless format family (screen-record, POV/hands, text-overlay) by what it has to prove, not by what's trending.
- A hook bank: multiple hooks per value×pain pair, each tied to a real app pain, tagged by cognitive trigger, held to a word-count rubric so it lands inside the first three seconds.
- Shot-listed 30–60s scripts — every shot names its visual source, on-screen text, duration, an audio type (never a copyrighted track), and a burned-in caption note. No talent, no face.
- Per-platform caption + hashtag SEO and a real CTA→install path: bio-hub trackable link, comment-pin assist, and message-match to your store listing.
- A dated 2–4 week calendar with one-shoot-three-platforms repurposing, plus a measure-and-iterate loop that tells you where the funnel leaks — the hook, the CTA, or the listing.
Who it's issued to: indie operators who want installs from video but have no interest in being on camera, and small teams who want a repeatable posting system instead of a new script every Sunday night.
Why not a free content prompt? A free prompt writes one clever script and stops. This is a system — app-tied hooks, faceless formats, a calendar, and a compliance pass (the TikTok disclosure toggle alone saves you a silent reach throttle). It's also honest about the part the hype merchants won't say: organic short-form is a grind, reach is falling, and nobody can promise you a viral hit. What it promises is many good at-bats and a way to learn from them. It's versioned and maintained as the platforms move their limits, and the field report shows a full run before you pay.
Runs on Claude Code, claude.ai, or any agent that reads markdown. No camera, no course.
FIELD REPORT real output, not a promise
Setup: the engine run end-to-end (SKILL.md Steps 0–7) for Dripwise, a fictional plant-care app — photo plant ID plus a personalized watering/fertilizing schedule — targeting anxious new plant parents who keep killing their plants, at 5 posts/week across TikTok/Reels/Shorts.
Honest frame, stated up front (the skill requires it): this is a repeatable system for many good at-bats, not a virality guarantee. Reach is falling industry-wide, reliable signal takes ~60–90 days of consistent posting, and no per-video reach or install number is promised.
Step 1 — Value ⇄ Pain matrix
| Value prop | Pain it kills | Trigger moment |
|---|---|---|
| Snap a photo → instant plant ID + care profile | Don't know what the plant is/needs | Bring home an untagged plant, zero info |
| Personalized watering schedule (species+light+pot+season) | Forget to water | Day 9, can't remember the last watering |
| Overwater/underwater symptom check-in | Overwater/underwater confusion | Leaves yellowing/drooping, no idea which way to correct |
| Personalized fertilizing schedule | Don't know what it needs (feeding) | Plant's alive but not growing, no idea why |
| Per-plant push reminders | Forget to water (at scale) | 6 plants, 6 needs, can't track any of them |
Step 2 — Content pillars (format-fit rule applied)
- Demo/Proof (proves it works) → Screen-record, Silent-satisfying
- Pain-POV ("that's me") → POV/hands+phone, Text-listicle
- Myth/Objection (corrects a wrong belief) → Myth-bust, React-over-screenshot
- Outcome/Transformation (before→after) → Before/after, Tutorial
Step 3 — Hook bank (excerpt: the 3 strongest of 15)
The full pass produced 15 hooks — all ≤14 spoken words / ≤7 on-screen words, ≥3 trigger types per value×pain pair. The three the summary flags as strongest:
- #7 "POV: you find your plant on day 11 of no water." — on-screen: "day 11. no water. oops" (Pain-POV)
- #11 "Yellow leaves don't mean what you think they mean." — "yellow leaves ≠ what you think" (Myth/Objection)
- #4 "One photo told me more than 3 months of Googling did." — "1 photo > 3 months of googling" (Demo/Proof)
Step 4 — One shot-listed faceless script (of three produced)
Script 1 — Pain-POV · POV/hands+phone · TikTok · ~27s · Hook #7
| # | Time | Visual source | On-screen text | VO |
|---|---|---|---|---|
| 1 | 0–2s | POV: hand pulls a droopy, forgotten plant from a corner, crispy leaves | day 11. no water. | "You find the plant you forgot existed." |
| 2 | 2–5s | POV: hand touches bone-dry soil | still alive?? | "Somehow it's still alive. Somehow." |
| 3 | 5–10s | Screen-record: open Dripwise, tap +, photograph the plant | scan it once | "So I scan it—" |
| 4 | 10–16s | Screen-record: app IDs "Pothos," generates care card + watering countdown | tells you what it needs | "—and it tells me exactly what this thing needs. Not a guess. A schedule." |
| 5 | 16–22s | Screen-record: lock-screen notification "Water your Pothos today" | reminder for THIS plant | "Then it just reminds me. For this exact plant." |
| 6 | 22–27s | Same plant weeks later, new leaf, healthy | [Dripwise] — link in bio | "It's called Dripwise. My plants stopped dying by accident." |
Audio: low-stakes relatable-comedy beat. Loop: "day 11" ↔ "stopped dying by accident." Disclosure: ON.
Step 6 — Compliance pass
Disclosure ON all 5 posts · real Dripwise UI only, no mocked results · no virality/earnings claims present · music = track-type only, no named songs · native re-export per platform, no watermarks.
Step 7 — Measurement
KPI ladder: 3s hold → completion% → profile visits → link taps → installs. Decision rules: weak completion (<50%) = rewrite the hook; good completion but low taps = sharpen the CTA; good taps but low installs = it's the listing, not the video. Judge on 9–12 posts (~2–3 weeks) per pillar before deciding, then double down on the top ~20% by completion+saves pattern — not the one exact video.
The honest closing call
The biggest risk to installs isn't the videos — it's that the App Store listing is still a placeholder. Until its first screenshot and subtitle literally say "photo ID + personalized watering schedule," every post pre-sells a click that lands on an unconfirmed promise. The engine flags this and hands off to ASO; it doesn't paper over it.
SERVICE RECORD living gear — updated as the factory learns
v1.0.0 — 2026-07-17
First issue. Authored from scratch: the end-to-end SOP (intake → value⇄pain matrix → pillars → hook bank → shot-listed faceless scripts → caption/CTA path → dated calendar → measure-and-iterate loop), the eight faceless format families, the hook-bank rubric, three reference cheatsheets, and worked mini-examples. Platform defaults stamped current as of 2026-07 with a verify-the-live-doc instruction so the numbers age gracefully. No invented metrics, no virality promises.
Every update ships free to owners — your locker always serves the latest version.
QUESTIONS
Does the agent shoot or post the videos for me?›
No. It authors the ready-to-shoot package — pillars, hook bank, shot-listed scripts, captions, and a dated calendar. You (or an editor) shoot and post from it. The skill doesn't shoot, edit, or auto-post anything.
Will this make my videos go viral?›
No, and anything that promises that is selling hype. Organic short-form is a grind — reach is falling and outcomes are power-law. The system manufactures many good at-bats and a way to learn from the retention data; reliable signal usually takes 60–90 days of consistent posting.
Do I really need no camera?›
Correct — every format is faceless by design: screen-record, POV/hands, text-overlay. No on-camera talent required. That's the point, not a limitation.
What does it need from me, and which agents does it run on?›
It needs your value props, ICP, and top user pains up front — without them it stops and asks, because pain-tied hooks are the whole differentiator. It runs on Claude Code, claude.ai Projects, Codex, or any agent that reads markdown; the reference files travel with it.