An ecom brand needs hundreds of native videos a month to matter in the feed. No human content team scales to that — but a content system does. This is ours: the Claude stack we run at UGC Ninja to feed creator campaigns for D2C brands. It turns one product into 30 angles, 100 hooks, native creator briefs, product stories that don't read like ads — and tells you why one video broke out while its twin died.
How to use it: every prompt below is standalone. Paste into Claude (or GPT — they all run on any modern LLM), replace the bracketed [INPUT] blocks with your product / script / data, run. The agents are the roles; the workflows are the order. You don't need all 15 every time — pick the workflow for the bottleneck you're staring at.
One rule before you start: the output is scaffolding, not replacement. If a prompt's output disagrees with what your operator gut says about your product — trust the gut, regenerate with better inputs.
═══
Each agent is a role with a job. In practice, an "agent" = a prompt (or two) run with a consistent persona. Set them up as Claude Projects / custom instructions, or just run the prompts raw.
| # | Agent | Job | Runs prompts |
|---|---|---|---|
| 1 | Angle Strategist | explodes one product into 30 angles, kills the weak ones | 1, 2 |
| 2 | Concept Briefer | turns a surviving angle into a shootable concept | 3 |
| 3 | Hook Writer | generates 100 hook variants per product | 4 |
| 4 | Hook Judge | filters 100 hooks down to the 10 that survive the feed | 5 |
| 5 | Scriptwriter | writes scripts creators actually want to shoot | 6, 11 |
| 6 | Brief Builder | writes the one-page native creator brief | 7 |
| 7 | Native Editor | detects ad-shape before it ships; enforces the do-not list | 8 |
| 8 | Casting Matcher | matches concepts to the right creators | 9 |
| 9 | Story Miner | digs the true product story + mines reviews for angles | 10, 12 |
| 10 | Viral Analyst | diagnoses breakouts vs dead twins, extracts templates, multiplies winners | 13, 14, 15 |