A multi-agent AI pipeline that researches your market, then writes production-ready launch content — scripts, threads, ads, landing pages, and Product Hunt launches.
Every piece of content goes through research, iterative writing, and a quality gate before it ships.
- You describe your product — brand, audience, tone, content type
- AI agents research your market — YouTube and X/Twitter research run in parallel
- Specialized agents write each section — hooks, body, CTA — each building on the last
- Weapons Check scores every line — Novelty + Copy Intensity must hit 9+/10 or it gets rewritten
- Final content assembles into a ready-to-use markdown file
The whole pipeline streams live in your browser so you can watch each agent work.
| Type | Length Options |
|---|---|
| Launch Video Script | 30s, 60s, 90s, 2min |
| Twitter/X Thread | 5 tweets, 10 tweets, 15 tweets |
| YouTube Ad Script | 15s, 30s, 60s |
| Landing Page Hero | Short, Long |
| Product Hunt Launch | Standard |
- Node.js v18+
- Anthropic API key or an active Claude Code session
git clone https://github.com/sinaghodsi93/content-king.git
cd content-king
npm installSet your API key (if not using Claude Code CLI auth):
export ANTHROPIC_API_KEY=your-key-herenpm run devOpen http://localhost:3111 in your browser.
- Fill in your product brief (brand name, what it does, audience, features, tone)
- Pick a content type and length
- Hit Run — the pipeline streams each agent's work live
- Grab your finished content from the output panel
| Step | Agent | What It Does |
|---|---|---|
| 00 | Brief | Validates your input and sets word budgets |
| 01 | YouTube Research | Finds what's working on YouTube in your niche |
| 02 | X Research | Finds what's landing on X/Twitter in your niche |
| 03 | Hook | Writes 4 hook options, iterates 3x, scores on 5 dimensions |
| 04 | Body | Writes the core content using research ammunition |
| 05 | CTA | Writes 2 close options (direct + soft), scored on 4 dimensions |
| 06 | Weapons Check | Scores every line — rewrites anything below 9/10 |
| 07 | Assemble | Builds final output file with metadata |
Research steps are automatically skipped when not relevant (e.g., YouTube research is skipped for Twitter threads).
pipeline-gui/
├── server.ts # Express backend + Claude Agent SDK integration
├── index.html # Browser UI (dark theme, real-time streaming)
├── pipeline/ # Agent prompt files (numbered for ordering)
│ ├── 00-brief.md
│ ├── 01-research-youtube.md
│ ├── 02-research-x.md
│ ├── 03-hook.md
│ ├── 04-body.md
│ ├── 05-cta.md
│ ├── 06-weapons-check.md
│ └── 07-assemble.md
├── output/ # Generated content files
├── workspace/ # Session working directories (gitignored)
├── brief.md # Current product brief
└── package.json
| Env Variable | Default | Description |
|---|---|---|
ANTHROPIC_API_KEY |
— | Your Anthropic API key |
PIPELINE_DIR |
./pipeline |
Directory containing agent .md files |
WORKING_DIR |
project root | Directory Claude Code operates in |
Each step is a markdown file in pipeline/. You can:
- Edit existing agents — modify the prompts to change writing style, scoring criteria, or iteration count
- Add new steps — create a new numbered
.mdfile (e.g.,03b-tone-check.md) and it appears in the GUI automatically - Remove steps — delete or rename the file
The server picks up all .md files in the pipeline directory, sorted by filename.
- Claude Agent SDK — AI agent execution
- Express — Server + SSE streaming
- Vanilla HTML/CSS/JS — No framework, no build step
MIT