Content Agent — AI Research → Reel Script Pipeline

A multi-agent system that researches trending topics in AI, Tech, and Startups, writes short-form video scripts (30s, 60s, 2min reels), and iteratively edits them for clarity, conversational tone, and viral hooks.

Architecture

┌─────────────────┐     ┌──────────────────┐     ┌─────────────────┐
│  Research Agent  │────▶│ Scriptwriter Agent│────▶│  Editor Agent   │
│                  │     │                  │     │                  │
│ • Topic discovery│     │ • 30s scripts    │     │ • Clarity pass   │
│ • Deep research  │     │ • 60s scripts    │     │ • Tone tuning    │
│ • Data & stats   │     │ • 2min scripts   │     │ • Hook optimizer │
└─────────────────┘     └──────────────────┘     └─────────────────┘

Agent Roles

Agent	Purpose
Research Agent	Discovers trending topics and produces detailed research briefs with stats, narratives, and surprising angles
Scriptwriter Agent	Converts research into multiple script lengths with hooks, visual cues, and captions
Editor Agent	Iterates each script through multiple passes, scoring hook power (1-10), clarity (1-10), and conversational tone (1-10) until approved or max passes reached

Setup

# Clone and install
pip install -r requirements.txt

# Configure your LLM provider
cp .env.example .env
# Edit .env with your API key

Usage

Full pipeline (auto-discover topics)

python main.py run --category ai --count 3

Full pipeline (specific topics)

python main.py run --topics "GPT-5 rumors, AI agents replacing SaaS, YC W25 batch trends"

Single topic deep-dive

python main.py single "OpenAI's new reasoning model"

Discover topics only (no scripts)

python main.py discover --category startups --count 5

Options

Flag	Default	Description
--category	all	ai, tech, startups, or all
--count	3	Number of topics to discover
--topics	—	Comma-separated topics (skips discovery)
--variations	3	Script variations per topic
--edit-passes	2	Max editing iterations per script
--output	output/	Output directory

Output Structure

Each run creates a timestamped directory:

output/
└── 20250101_143022/
    ├── research/          # Research briefs per topic
    ├── scripts_draft/     # Raw scripts from Scriptwriter
    ├── scripts_final/     # Polished scripts from Editor
    │   └── 1_topic_name/
    │       ├── 30s.md
    │       ├── 60s.md
    │       └── 2min.md
    └── summary.json       # Structured run metadata

Supported LLM Providers

Set LLM_PROVIDER in .env:

• openai — GPT-4o, GPT-4-turbo, etc.
• anthropic — Claude Sonnet, Claude Opus, etc.