$cd ../use-cases/
π€ Multi-AgentAdvanced45 min setup
$ cat multi-agent-content-factory.md
const ContentFactory = {agents: 3}
/** One idea β researched outline β draft β thumbnail β scheduled posts. All inside OpenClaw. */
π Pipeline Results
3
specialized agents
70%
faster content production
0
copy-paste between tools
24/7
autonomous research loop
how_it_works.md
βοΈ How the Multi-Agent Content Factory Works
1.
Research Agent scans sources
Runs on a continuous loop monitoring RSS feeds, Reddit threads, Hacker News top stories, and YouTube transcripts for trending topics in your niche. It scores each candidate by engagement velocity and semantic relevance to your content strategy, then generates a structured research brief with primary sources, competitor angles, and a suggested headline list.
2.
Writer Agent drafts the piece
Receives the research brief and writes a full blog post, newsletter issue, or tweet thread calibrated to your defined brand voice. It respects your configured reading level, tone (authoritative vs. conversational), and target word count. Multiple headline variants are generated for A/B testing, and a meta description is included for SEO.
3.
Designer/Publisher Agent packages it
Generates detailed image prompts for thumbnail creation via Midjourney or DALL-E, applies your brand color palette and typography guidelines, then schedules the social media cross-posts across Twitter/X, LinkedIn, and Mastodon at the optimal engagement windows for your audience timezone.
4.
Human-in-the-loop approval
You receive the complete package in your Telegram or Discord review channel β draft, thumbnail preview, and scheduled posts. Reply 'approve' to trigger publishing, or add revision notes and the Writer Agent automatically incorporates feedback and re-submits. The loop completes in under 5 minutes of your time per piece.
setup.yaml
π¦ Build It in 45 Minutes
// Step 1: Set up a Discord server with channels for each agent
#research-briefs
#drafts-pending
#approved-publish
#content-archive
// Step 2: Configure three OpenClaw agent personas
agents:
researcher:
role: "Research trends and produce briefs"
channel: "#research-briefs"
writer:
role: "Write articles from briefs"
channel: "#drafts-pending"
publisher:
role: "Schedule posts and generate thumbnails"
channel: "#approved-publish"
// Step 3: Connect skills for search, browser, and publishing
skills:
- brave_search
- browser_automation
- image_generation
- ghost_cms
- twitter
// Step 4: Start the loop
$ openclaw cron add --name content-factory --schedule '0 * * * *'
agent_architecture.tsv
ποΈ Agent Architecture
| Agent | Input | Output | Channel |
|---|---|---|---|
| Researcher | RSS / Reddit / Search | Structured brief with sources | #research-briefs |
| Writer | Brief | Draft + headline options | #drafts-pending |
| Editor (you) | Draft | Approved / revision request | DM or thread |
| Publisher | Approved draft | Published post + thumbnail | #approved-publish |
β Multi-Agent FAQ
Q1. Do the agents talk to each other directly?
They communicate through shared channels and a structured memory store. Each agent reads the previous agent's output as a formatted artifact in the channel, then produces the next stage. You can also configure explicit handoff triggers β for example, the Writer Agent only fires when a research brief contains at least 5 sources and a confidence score above 0.8.
Q2. Can I use this for video scripts?
Yes. Replace the Writer Agent persona with a 'scriptwriter' role that outputs hook, script segments, and B-roll suggestions instead of prose. Add a YouTube skill for metadata generation (titles, descriptions, tags) and thumbnail prompt creation. The same approval loop works identically for video content.
Q3. How do I prevent agents from running unsupervised?
Use the human-in-the-loop gate built into the config. The Publisher Agent only fires when you reply 'approve' in the review channel. You can also set a daily content quota β even if the Research Agent surfaces 20 trending topics, it will only brief the Writer Agent on the top-scored 3 without your explicit expansion.
Q4. What channels work best?
Discord is ideal because you can create dedicated channels per content stage (#research-briefs, #drafts-pending, #approved-publish) and archive everything automatically. Telegram works well for smaller teams or solo creators. Slack is supported via the webhook skill but lacks Discord's channel-as-memory architecture.
Q5. How does the Research Agent know what's relevant to my niche?
You define a niche description and a list of seed keywords in the agent config. The Research Agent scores each candidate topic against these seeds using semantic similarity. You can also provide negative keywords to filter out irrelevant content categories.
Q6. Can the Writer Agent match my existing writing style?
Yes β provide 3β5 sample articles in the agent's context as style examples. The Writer Agent uses these as few-shot style references, mirroring your sentence structure, vocabulary density, section patterns, and preferred heading style. The match quality improves as you add more samples.
Q7. What CMS and publishing platforms are supported?
Ghost, WordPress (via REST API), Webflow (via CMS API), and Notion are natively supported. Newsletter platforms include Beehiiv, ConvertKit, and Mailchimp via their APIs. Social scheduling targets Twitter/X, LinkedIn, and Mastodon. Additional platforms can be added as custom webhook skills.