# Agentic Pipeline for GTA 6 Assets

> A 4-step pipeline to produce animated assets with agents: brief, build, render, quality check. One step, one file, one agent.

Status: metodo · Updated: 2026-11-18

**TL;DR:** A 4-step pipeline: brief, build, render, QA. One agent per step, one file per step, independent QA.

Agentic work holds up when it is **progressive**: every step produces a file, no step depends on one giant agent.

## The 4 steps

| Step | Input | Output |
|---|---|---|
| 1. Brief | Prompt Lab | `brief.md` |
| 2. Build | `brief.md` | `asset.html` |
| 3. Render | `asset.html` | `out.mp4` or `out.gif` |
| 4. QA | all files | `qa.md` |

## Rules

- **One agent per step**, at most two in parallel. Never many agents on big tasks: if a limit trips, nothing is left.
- **Save at every step**: the result lives in a file, not in the conversation.
- **Independent QA**: the checker is not the builder. Hand it the checklist, not your opinion.

## Minimum QA checklist

- The loop does not jump at the end
- Works with `prefers-reduced-motion`
- No external resources loaded
- Weight within the limit
- No third-party material (logos, fonts, audio, characters)

For reusable style rules see the [skill](/en/guide/studio-claude-code-skill-assets).

## FAQ

### How do you organize AI agents to produce assets?
Four steps (brief, build, render, QA), one agent and one file per step, at most two agents in parallel and an independent quality check.
