# Absorbed skill: social-shorts-growth-engine

---
name: social-shorts-growth-engine
description: Plan and build automated short-form video growth pipelines for YouTube Shorts, TikTok, Instagram Reels, and similar channels with retention-led ideation, human approval, and analytics feedback.
version: 1.0.0
---

# Social Shorts Growth Engine

Use this skill when building or improving systems that generate, review, render, or publish short-form social videos, especially automated/faceless pipelines for YouTube Shorts, TikTok, Instagram Reels, or cross-platform repurposing.

## Core principle

Do **not** treat the project as merely a video generator. Treat it as a growth engine:

1. Research what the platform currently rewards.
2. Generate hypotheses for hooks/formats/niches.
3. Score ideas before rendering.
4. Produce reviewable drafts/previews.
5. Require human approval before posting.
6. Capture analytics and feed winners back into the next batch.

## Default workflow

1. **Load project constraints**
   - Channel/topic/session boundaries.
   - Existing code locations and repo.
   - Secret-handling rules.
   - User’s desired niches and no-go topics.

2. **Research current platform rules**
   - Prefer official/near-official sources: YouTube Help, TikTok Creator Academy/Terms, Meta docs.
   - Record only durable rules in a concise reference file, not full scraped pages.
   - Verify monetization details before recommending video lengths.

3. **Define content pillars**
   - Keep pillars distinct enough to measure.
   - Avoid mixing sensitive identity topics with AI/automation unless the user explicitly wants that brand strategy.
   - If the user says a pillar is “heikel”/sensitive, pause it in config rather than deleting it, with a reason and revisit condition.

4. **Build a test batch**
   - Start with 6–12 ideas, not dozens of rendered videos.
   - Include a mix of growth candidates and monetization-length candidates.
   - For TikTok Rewards experiments, include selected 61–75s candidates only after hook strength is high.
   - For growth, prefer 25–45s candidates.

5. **Score before rendering**
   Score ideas on:
   - direct self-relevance in first 1–2 seconds
   - open loop
   - specificity
   - tension/contrarian angle
   - concrete payoff
   - visual pattern interrupt
   - comment/conversation trigger
   - niche authority
   - policy/brand risk

6. **Generate Telegram/human review packages**
   Each draft/preview should include:
   - title
   - duration
   - platform targets
   - hook
   - script draft or final voiceover
   - visual concept
   - caption/hashtags
   - quality/retention score
   - risks/policy notes
   - approval commands

7. **Do not auto-post until approved**
   Use explicit approval commands such as:
   - `Freigabe alle`
   - `Freigabe YouTube`
   - `Freigabe TikTok`
   - `Freigabe Instagram`
   - `Ablehnen`
   - `Ändern: ...`

8. **Analytics loop**
   After posting, track at least:
   - platform
   - pillar/niche
   - format
   - hook
   - duration
   - views after 2h/24h/7d
   - likes/comments/shares/follows
   - retention/average watch time if available

## Retention heuristics

People tend to stop scrolling for:

- **Immediate self-relevance:** “If your brain has 47 tabs open…”
- **Open loops:** “do this before…”, “why…”, “stop…”, “not…”
- **Specificity:** numbers, named tools/platforms, concrete situations.
- **Tension:** challenges a common belief without cheap outrage.
- **Concrete payoff:** template, prompt, before/after, exact step.
- **Visual pattern interrupt:** movement/change in the first seconds, not static AI art.
- **Conversation trigger:** specific question or keyword CTA.
- **Niche authority:** a recognisable point of view, not generic advice.

Reject or revise drafts with:

- “In this video…” slow intros.
- generic “4 tips” without a fresh angle.
- pure slideshow/static visuals.
- abstract AI backgrounds that do not communicate the hook.
- AI-gibberish UI/text artifacts.
- broad advice with no visible payoff.

## Monetization notes

Always re-check current official docs before major platform decisions. Current durable operating assumptions:

- YouTube Shorts can be vertical/square and up to 3 minutes.
- YouTube Shorts YPP path has historically required 1,000 subscribers plus large valid Shorts-view thresholds in a recent period.
- TikTok Creator Rewards has historically rewarded original, high-quality videos longer than 1 minute and required account eligibility such as follower/view thresholds and good standing in selected countries.
- TikTok’s reward-eligible content should avoid low-quality patterns: copied/minimally altered work, meaningless reactions, blurry visuals, pure slideshows, clickbait without value, and policy-risky content.

See `references/retention-and-monetization-notes.md` for concise notes from the AutoShortsBot session.

## Implementation patterns

For repositories that generate shorts:

- Keep secrets, generated media, `.env`, and credentials out of Git.
- Store test batches as structured data under `data/drafts/`.
- Store human-review drafts under `data/reviews/` if they are lightweight text.
- Add validation tests for batch shape and required fields.
- Add heuristic retention scoring before rendering.
- Treat rendering as later-stage; validate scripts and hooks first.

## Pitfalls

- Do not immediately render all ideas. First validate hooks, scripts, and review format.
- Do not assume a niche is safe because it is popular. Sensitive topics can confuse brand positioning when mixed with AI/automation.
- Do not rely on platform tags/hashtags as the growth strategy. Hooks, retention, satisfaction, and niche fit matter more.
- Do not post automatically from an agent without explicit approval and a side-effect boundary.
- Do not print or commit tokens from Google Drive or local secret files.