Content Pipeline

The Content Pipeline is a multi-agent workflow that automates the full lifecycle of content creation — from research and topic selection through drafting, review, and publishing. It uses the Clop Cabinet squad to coordinate specialized agents through a gated approval process.

Pipeline Overview

The content pipeline follows eight stages, each handled by a specialized agent:

Research → Topic Selection → Draft → Media → Hashtags → Review → Publish → Report
   │           │              │       │         │          │         │         │
 Mimir      Frigg          Bragi   Bragi     Bragi     Daniel    Bragi     Loki

Stages in detail

Stage	Agent	Description
1. Research	Mimir	Scans RSS feeds, news, trends, and bookmarks for relevant topics
2. Topic Selection	Frigg	Reviews research, selects the best topic, defines angle and audience
3. Draft	Bragi	Writes the content piece following platform-specific guidelines
4. Media	Bragi	Generates or selects images, creates carousel slides if needed
5. Hashtags	Bragi	Researches and adds platform-optimized hashtags and CTAs
6. Review	Daniel (human)	Approval gate — human reviews draft before publishing
7. Publish	Bragi	Posts to the target platform via API
8. Report	Loki	Monitors engagement metrics and generates a performance report

Agent Roles

Mimir — Knowledge Curator

Mimir is the research engine. It has access to:

RSS feed readers and blog watchers
Link transcription (YouTube, articles)
Bookmark management (Linkding)
Web search
PDF summarization

Mimir outputs structured research briefs with sources, key insights, and suggested angles.

Bragi — Content Creator

Bragi handles all content creation. It knows:

Platform-specific formatting (LinkedIn, Twitter, Instagram, blog)
CTA templates and engagement patterns
Hashtag research and optimization
Image prompt generation
Post scheduling

Bragi works from templates stored in its workspace:

squads/clop-cabinet/cma/
├── CHECKLIST-POST-LINKEDIN.md
├── CTA-TEMPLATES.md
├── POST-TEMPLATE.md
├── PLAYBOOK-LINKEDIN.md
├── drafts/
├── images/
├── posts/
└── scripts/

Loki — Analytics Monitor

Loki tracks post-publication metrics:

Engagement rates (likes, comments, shares)
Impression counts
Follower growth
Content performance comparisons
Weekly and monthly trend reports

Frigg — Coordinator

Frigg orchestrates the entire pipeline:

Routes research requests to Mimir
Assigns drafting tasks to Bragi
Manages the approval queue
Sends drafts to Daniel for review
Triggers publishing on approval

Starting a Pipeline

Via Mission Control

Open Mission Control at http://localhost:3000
Navigate to Pipeline in the sidebar
Click New Pipeline
Configure:
- Platform: LinkedIn, Twitter, Instagram, Blog, Newsletter
- Topic: (optional) Let Mimir research, or provide a topic
- Tone: Professional, casual, technical, storytelling
- Target date: When to publish
Click Start — the pipeline begins at the Research stage

Via agent message

Send a message to Frigg (the coordinator):

openclaw agent --agent frigg -m "Start a content pipeline for LinkedIn. Topic: AI agents in software development. Target: tomorrow 10am."

Via cron (automated)

Set up a recurring content pipeline:

openclaw cron add \
  --name "weekly-linkedin-post" \
  --agent frigg \
  --cron "0 8 * * 1" \
  --message "Start a content pipeline for LinkedIn. Research trending AI/dev topics and draft a post for today." \
  --session isolated \
  --model sonnet

The Gate System

Gates are human approval points that prevent automated publishing without review. This is a core safety feature of ClawHalla.

How gates work

An agent completes its stage and moves the task to Review
Frigg creates an Approval Request via the approvals API
Daniel receives a notification (Telegram, Discord, or MC)
Daniel reviews the draft and responds:
- Approve: “aprovado”, “pode enviar”, “go” — pipeline continues
- Adjust: “change the title to…” — Bragi revises and re-submits
- Reject: “rejeitar”, “cancelar” — pipeline stops

Gate configuration

Gates are enforced by the approval system in SOUL.md:

## Approval Law -- Inviolable

Daniel approves everything. No exceptions.

Before any external action:
1. Complete the work
2. Move to Review on the board
3. Notify Daniel with the draft
4. Wait for explicit approval
5. Only then execute

Implicit approval does not exist.
Silence is not approval.

Viewing pending approvals

Mission Control: Navigate to Approvals in the sidebar.

API:

curl http://localhost:3000/api/approvals

Response:

{
  "pending": [
    {
      "id": "apr_m1x2y3",
      "taskId": "LinkedIn post: AI agents in dev",
      "requestedBy": "bragi",
      "approver": "daniel",
      "status": "pending",
      "command": "publish",
      "reason": "Draft ready for review. 1200 words, 3 images.",
      "createdAt": "2026-03-25T10:00:00Z"
    }
  ],
  "history": []
}

Platform Support

Full integration through the LinkedIn API:

Text posts with rich formatting
Image and carousel posts
Comment monitoring and engagement
Connection request management
Analytics tracking

Configure LinkedIn in Mission Control Settings by adding your LINKEDIN_ACCESS_TOKEN to the vault.

Twitter/X

Post creation, thread support, and engagement monitoring.

Instagram

Image posts with captions and hashtags. Requires a connected business account.

Blog

Generate markdown articles for your blog (Astro, Hugo, Next.js, etc.). Output is saved to a configured directory.

Generate newsletter content compatible with email platforms (Mailchimp, ConvertKit, Resend).

Scheduling and Automation

Manual scheduling

Set a publish date when creating a pipeline:

openclaw agent --agent frigg -m "Draft a LinkedIn post about ClawHalla v1.0. Schedule for Thursday 10am BRT."

Automated weekly pipeline

Use cron to create a fully automated content cadence:

# Monday: Research trending topics
openclaw cron add --name "monday-research" --agent mimir \
  --cron "0 8 * * 1" --session isolated \
  --message "Research trending AI and dev topics. Save findings to workspace."

# Tuesday: Draft LinkedIn post
openclaw cron add --name "tuesday-draft" --agent bragi \
  --cron "0 9 * * 2" --session isolated \
  --message "Read Mimir's latest research. Draft a LinkedIn post. Submit for review."

# Wednesday: Check approval status
openclaw cron add --name "wednesday-check" --agent frigg \
  --cron "0 10 * * 3" --session isolated \
  --message "Check if any content is approved and pending publication. Publish if approved."

Monitoring pipeline status

Check the current state of all pipelines:

curl http://localhost:3000/api/tasks?status=in_progress

Or check the board directly:

curl http://localhost:3000/api/board/sync?project=clawhalla | jq '.tasks[] | select(.tags | contains("content"))'

Best Practices

Always review before publishing. The gate system exists for a reason. Automated content without human review can damage your brand.
Feed Mimir quality sources. The research stage is only as good as its inputs. Configure RSS feeds, bookmarks, and topic lists relevant to your audience.
Iterate on templates. Bragi’s output quality improves dramatically when you provide good templates in POST-TEMPLATE.md and CTA-TEMPLATES.md.
Track what works. Use Loki’s analytics reports to understand which topics, formats, and posting times perform best. Feed this back into Mimir’s research priorities.
Start with one platform. Get the LinkedIn pipeline working well before expanding to other platforms. Each platform has different formatting rules and audience expectations.