# Content Dna

> Reverse-engineer any TikTok or Instagram video into a 7-dimension Content DNA breakdown using Gemini Flash 3. Identifies topic, angle, hook structure, story structure, visual format, key visuals, and audio. Also isolates portable bricks and generates 3 adapted concepts for your niche. Use when: user wants to analyze a video, understand why something went viral, steal a format, or generate adapted content concepts from an outlier.

- Skill: `themattberman/content-dna` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add themattberman/content-dna`
- Raw SKILL.md: https://api.skillmd.com/api/skills/themattberman/content-dna/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: TheMattBerman (https://skillmd.com/u/themattberman)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/themattberman/content-dna

---


# Content DNA

Reverse-engineer what made a video work. Seven dimensions, structured output, portable bricks you can adapt.

## Usage

### Single video
```
/content-dna https://www.tiktok.com/@creator/video/123
```

### With adaptation for your niche
```
/content-dna https://www.tiktok.com/@creator/video/123 --adapt --niche "AI marketing"
```

### With explicit voice choice
```
/content-dna https://www.tiktok.com/@creator/video/123 --adapt --niche "pet brand" --voice-mode analysis
/content-dna https://www.tiktok.com/@creator/video/123 --adapt --niche "pet brand" --voice-mode personal
/content-dna https://www.tiktok.com/@creator/video/123 --adapt --niche "pet brand" --brand /path/to/client-voice.md --voice-mode brand
```

### Shell
```bash
bash ~/clawd/skills/content-dna/scripts/content-dna.sh \
  --url "VIDEO_URL" \
  --adapt \
  --niche "AI marketing"
```

## What You Get

**7 Dimensions:**
1. **Topic** — what the video is about at a high level
2. **Angle** — the specific take that makes it unique
3. **Hook Structure** — visual hook + text on screen + spoken opening line
4. **Story Structure** — beat-by-beat narrative
5. **Visual Format** — talking head / split-screen / green screen / etc
6. **Key Visuals** — the specific shots that carry the idea
7. **Audio** — VO style, pacing, music, sound design

**Performance Layer:**
- Primary driver + secondary driver
- Engagement type: debate / utility / inspiration / entertainment / FOMO
- Why comments were high

**Strategic Readouts:**
- Brick name
- Brick definition
- One-paragraph explanation of why it worked
- Plain-English takeaway: `Use this when...`

**Portable Bricks:**
- The reusable mechanical elements (rated high/medium/low portability)
- Why each transfers to other niches

**Adapted Concepts (with --adapt):**
- 3 ready-to-brief concepts for your niche
- Each uses a different portable brick
- Written in the selected output voice
- Keep source niche and output voice separate when needed

## How It Works

1. ScrapeCreators fetches video metadata + transcript (if available)
2. Cover image + metadata sent to Gemini Flash 3 via OpenRouter
3. Gemini returns structured JSON with the 7-dimension breakdown + performance layer + brick analysis + adapted concepts
4. The script prints a human-readable markdown-style breakdown for chat use and saves the enriched JSON to `~/clawd/workspace/content-dna/outputs/`

## Source Niche vs Output Voice

Treat these as separate choices:
- **Source niche**: where the outlier came from
- **Output voice**: who the adapted concept is for

Use this decision rule:
- If you want brand-native output, pass an explicit brand voice file.
- If you want to borrow mechanics from another niche for your own content, use personal voice intentionally.
- If you just want analysis, use `--voice-mode analysis` and avoid forcing a brand voice onto the adaptation.

If the niche and active voice look mismatched, ask:
> Do you want this adapted for the niche's brand voice, your own voice, or just as raw pattern analysis?

## Integration with Larger Pipelines

Content DNA should sit in the middle of the pipeline:
1. Discovery / signal pull
2. Content DNA analysis
3. Pattern aggregation
4. Final generation in the selected voice

If a user asks for the **actual / full / real** pipeline, do not skip the analysis stage and do not fill gaps with made-up output.