The LinkedIn Ghostwriter is a fully autonomous content pipeline. Once configured, it runs every week without any manual input — generating a post, sending it to Slack for review, and publishing after approval.
How It Works
A GitHub Actions cron job triggers agents/linkedin-ghostwriter/main.py every Monday morning. The agent reads the brand voice guide and content pillars from config, calls the Claude API to generate a post, then validates output against word count and tone constraints.
The draft lands in a Slack channel with three buttons: Approve, Edit, or Reject. Approving triggers the LinkedIn API to publish. Rejecting closes the loop. Editing opens a modal where the post can be revised before publishing.
Published posts are archived to content/published/ in the repository for historical tracking.
Architecture
- Claude API — Haiku for fast drafts, Sonnet for rewrites when tone misses the mark
- Slack SDK — Interactive message blocks with button actions routed through a webhook handler
- GitHub Actions — Scheduling, secret management, and workflow orchestration
- Tweepy / LinkedIn API — Cross-posting support for Twitter/X when enabled
Results
Consistent weekly publishing without the mental overhead of staring at a blank compose box. The posts stay on-brand because the system prompt encodes voice, not just topics.