# Absorbed skill: social-shorts-automation

---
name: social-shorts-automation
description: Build, audit, and improve automated YouTube Shorts/TikTok/Reels generation systems, including content strategy, hooks, rendering pipelines, publishing, and analytics feedback loops.
version: 1.0.0
author: Hermes Agent
license: MIT
metadata:
  hermes:
    tags: [social-media, shorts, tiktok, youtube-shorts, video-generation, growth, content-strategy, automation]
    related_skills: [youtube-content, xurl]
---

# Social Shorts Automation

Use this skill when the user wants to build, review, or improve a system that automatically creates or repurposes short-form vertical videos for TikTok, YouTube Shorts, Instagram Reels, or similar platforms.

## When to load

- User asks about automated creation of social-media shorts.
- User provides or references scripts for video generation, clipping, captioning, TTS, image generation, publishing, or analytics.
- User wants to grow followers, improve views, monetize a channel, or brainstorm short-form content formats.
- User asks for FaithTok, AI content, life optimisation/self-improvement shorts, or recurring short-form series.

## Core principle

Do **not** treat the product as merely a video renderer. Treat it as a **Shorts Growth Engine**:

1. idea/trend mining
2. hook generation
3. script generation
4. visual plan
5. quality gate
6. render
7. publishing package
8. analytics feedback loop
9. next-topic selection based on performance

A technically working renderer can still produce low views if hooks, identity, retention, and feedback loops are weak.

## Audit workflow for an existing shorts bot

1. **Inventory the pipeline**
   - script generation model/provider
   - image/video source
   - TTS/voice provider
   - subtitle/caption rendering
   - music source and licensing
   - output dimensions/fps/duration
   - metadata generation
   - publishing integration
   - analytics storage

2. **Inspect sample outputs**
   - Check duration with `ffprobe`.
   - Extract frames at early, middle, and late timestamps.
   - Review visually for hook strength, readability, AI artifacts, brand consistency, and mobile-safe framing.

3. **Review generated metadata**
   - Titles, descriptions, captions, hashtags.
   - Look for generic wording, weak promises, overlong descriptions, or mismatched platform tone.

4. **Diagnose growth blockers**
   - Generic topic selection.
   - Weak 0–2 second hook.
   - No human identity or repeated series format.
   - Overlong runtime for a new account.
   - AI-image artifacts or obvious “AI slop”.
   - No retention/publishing analytics loop.

5. **Recommend next system step**
   - Prefer small experiments: 3 niches × 3 formats × 3 hook styles before large refactors.
   - Add scoring and analytics before scaling generation volume.

## Recommended content-engine concepts

Represent each planned short as structured data:

- `niche`: e.g. `life_optimization`, `ai_systems`, `faithtok`
- `format`: e.g. `contrarian`, `confession`, `tutorial`, `prayer`, `story`, `list`, `warning`
- `duration_target`: e.g. 20, 35, 55 seconds
- `platform`: `tiktok`, `youtube_shorts`, `reels`
- `language`: `de`, `en`
- `hook`: exact first spoken/on-screen sentence
- `script`: final spoken script
- `visual_plan`: per-beat visuals
- `caption`: platform caption
- `cta`: comment/follow/save prompt
- `risk_notes`: medical/finance/religious/policy concerns if any

## Hook quality gate

Before rendering, score every script. Reject or revise if it fails:

- Hook appears in the first 1.5 seconds.
- Hook creates curiosity, tension, identity, confession, surprise, or clear pain.
- One clear idea only; no broad essay.
- First visual frame communicates a problem/person/contrast, not generic ambience.
- Script avoids generic advice viewers have heard 200 times.
- There is a concrete reason to comment, save, share, or follow.
- Ending can loop naturally into the opening when possible.

## Early-account duration guidance

For new or low-traction accounts, default to shorter experiments:

- 18–25s: aggressive hook and format tests.
- 30–45s: value shorts with one useful idea.
- 55–60s: only when the story or emotional arc is strong.
- Avoid drifting into 80–90s explainers unless analytics proves retention.

## Visual strategy

Prefer visuals that signal trust, humanity, and specificity:

- Human face, hands, desk, phone, Bible, notebook, screen recordings, real B-roll.
- Kinetic captions and large on-screen hook text in the first two seconds.
- Pattern interrupts every 2–3 seconds: zoom, cut, object, caption emphasis, scene change.
- Avoid AI-generated UI screens, fake app icons, fake text, illegible pseudo-typography, and generic abstract graphs unless intentionally stylistic.

## Analytics feedback loop

Store a small performance record per published video:

```json
{
  "video_id": "...",
  "platform": "youtube",
  "niche": "faithtok",
  "format": "confession",
  "hook": "...",
  "duration": 32,
  "published_at": "2026-05-22T20:00:00+02:00",
  "views_24h": 1200,
  "avg_retention": 0.61,
  "likes": 83,
  "comments": 12,
  "shares": 4,
  "follows": 9
}
```

Use analytics to decide the next batch: repeat winners, vary one variable at a time, and retire weak hooks/formats.

## Content positioning patterns

For life optimisation, AI, and FaithTok, stronger positioning comes from identity and contrast, not generic tips.

- Life optimisation: “systems, not motivation”; “less chaos in 60 seconds”.
- AI: practical AI for normal people; AI as servant, not identity.
- FaithTok: faith for overloaded modern life; emotionally honest, not kitschy or preachy.
- Hybrid positioning example: “Faith, focus, and AI systems for a better life.”

## Pitfalls

- Do not just generate more videos when views are weak; first improve hooks and add analytics.
- Do not let image generation produce text/UI elements unless the model is reliable; fake labels damage trust.
- Do not commit `.env`, tokens, generated private assets, or raw secrets to GitHub.
- Do not claim platform-specific monetization success without current analytics.
- For religious content, keep tone grounded, sincere, and non-manipulative; avoid cringe, overclaiming, or weaponized guilt.

## References

- `references/autoshortsbot-initial-audit.md` — project-specific initial audit notes and improvement direction for the AutoShortsBot prototype.