Prompt System Overview

# Prompt System Overview

The dashboard's prompts live inside `src/AgentOpsDashboard.jsx`. This note explains how
they're organized and what is intentionally not extracted into separate published files.

## Structure

Each step's prompt is composed from three layers:

1. **Base system prompt** — channel identity + platform constraints (TV-14 / SFW /
   platform-compliant).
2. **Creative DNA** — abstract craft principles per craft area (idea, script, shot list,
   voice, music, images, platform behavior). These are written as *principles*, not as
   copies of any specific show, character, or episode.
3. **Step instructions + output contract** — the concrete task and, for idea generation,
   a strict JSON schema the model must return.

Several steps also inject **memory context** (recently rejected titles, established
characters, taste history, tone preferences) so generation improves across sessions.

## What is intentionally NOT published separately

- I have not dumped the full proprietary prompt text into standalone files here. The
  prompts are visible in the source for transparency, but if this project is developed
  further, the most refined prompt assets may be kept private.
- No copyrighted titles/characters/episodes are embedded — the DNA is deliberately
  abstract. If you add anything more specific later, review it before publishing.

## If you extract prompts later

Keep them abstract, keep secrets out, and review for any third-party IP before committing.