Skip to main content
Here’s what most people do: open ChatGPT, paste a topic, get a draft, close the tab, do it again tomorrow with zero memory of what happened yesterday. That’s not a content pipeline. That’s a very expensive coin flip. An OpenClaw content pipeline is three persistent agents that run 24/7, build up knowledge of your brand over weeks and months, store your style guides and drafts on their own filesystems, and use real tools to research and produce content. Week 8 output is dramatically better than week 1 because the agents actually learn.

​
Why this actually works

They build knowledge over time. The research agent remembers every topic it’s covered. The writer learns which phrasing you approve and which you reject. The editor maintains a consistent standard across months of drafts. This compounds. It’s not starting from scratch every session. They work on their own. Drop a topic and walk away. The research agent autonomously pulls sources using web search, writes a structured brief to its filesystem, and is ready when you come back. No babysitting required. They have real filesystems. Style guides, brand voice docs, content calendars, past drafts. All of it lives on the instance’s volume. The agent reads them for every piece it produces. When it finishes a draft, it writes the file to disk so you have a persistent record. Not a chat message you have to scroll back through to find. They have real tools. Web search for current data and competitor analysis. File handling to read your style guide and write drafts. The agents pull live information and produce artifacts, not just text responses.

​
The setup

Create three instances with distinct roles: Research instance connects to Slack. Drop a topic, URL, or competitor link and the agent gets to work. It uses web search to pull current sources, cross-references against its filesystem of previously researched topics (stored in briefs/), identifies gaps, and writes a structured brief to disk. System prompt is tuned for analysis, not prose. Writer instance connects to Discord. Point it at a research brief (or paste one in) and it produces a first draft. Blog post, newsletter, social thread, whatever. It reads your brand voice guide and content guidelines from its filesystem on every single draft. Finished drafts go to drafts/ on disk. System prompt includes tone, formatting rules, and word count targets. Editor instance connects to Telegram for quick back-and-forth. Send it a draft and it returns line-level feedback: clarity, flow, factual claims to verify, SEO improvements. It reads your editorial standards from its filesystem and applies them consistently. It maintains a running log of common issues in feedback/edit-log.md so patterns get caught earlier over time.

​
How this compounds

Week 1. You drop five topics into the research Slack. The agent researches each one, writes briefs to briefs/, and you have a backlog ready to go. The writer produces drafts from those briefs, reads your style guide from disk, and writes output to drafts/. The editor gives feedback and logs patterns. Week 4. The research agent has covered 20 topics. When you drop a new one, it checks briefs/ and tells you it already covered a related angle three weeks ago. Suggests building on it instead of starting from scratch. The writer has seen dozens of your edits and produces drafts that need fewer passes. The editor’s log has identified your three most common issues and flags them upfront. Week 8. The pipeline is producing publish-ready content with one editing pass. Each agent has internalized your standards through persistent memory. New team members can drop topics into Slack and get quality output without learning the style guide themselves. The agents enforce it.

​
What to configure

​
Filesystem per instance

  • Research. briefs/ directory for output, sources/ for reference material, previous research the agent should know about
  • Writer. drafts/ for output, guides/brand-voice.md, guides/content-guidelines.md, example posts that represent your target quality
  • Editor. feedback/edit-log.md for pattern tracking, guides/editorial-standards.md, SEO checklist, house style reference

​
Skills

  • Web search is essential for the research instance and useful for fact-checking in the editor
  • File handling for all three instances to read from and write to their filesystems

​
Personas

  • Research. Analytical, structured output, no creative flourish. Knows how to use web search effectively and organize findings.
  • Writer. Creative within your brand voice constraints. Reads the style guide before every draft.
  • Editor. Strict, direct. Does not soften criticism. Maintains the running edit log.
Three instances fit on the Pro plan. Add localization instances per language on the Max plan to translate and culturally adapt content after editing.