Most people spend two to three hours reformatting one piece of content for different platforms. They write a post on LinkedIn. Then they try to rewrite it for Twitter. Then they try to design slides for Instagram. It takes too much time. When client work gets heavy, posting stops completely. I built a custom skill for Google Antigravity called content-repurposer. Here is how I ran it today on a 25-minute YouTube video by Stephen G. Pope about AI agents, and what came out of it. 1. The Input I gave the agent one simple command with the YouTube link: bash " /content-repurposer https://youtu.be/VXx2jU7ObTY " That was all. I did not copy the transcript manually. I did not write a long custom prompt. 2. What Happened In 30 Seconds The agent ran in the background and: Extracted the full spoken audio transcript (4,748 words). Identified the core technical lesson: most developers build toy agents that read from local disk, but production systems require an Agent File System (AFS) with remote APIs, pre-warmed Git pools, and background conflict fixers. Generated three distinct hook angles to test. Translated the lesson into platform-native posts adhering to each algorithm's specific rules. 3. What I Got Out Of It From that single YouTube link, the skill generated: 3 Hook Angles to Test:- Contrarian: "Why 90% of AI agents fail in production." Data / Numbers: "How to run 25+ agents simultaneously with zero merge conflicts." Personal Story: "What broke when I first deployed multi-agent pipelines to the cloud." Twitter/X Thread (7 Tweets):- 100% link-free Tweet 1 (to avoid the algorithm link penalty). Single-line spacing and emoji bullets. Bookmark-optimized call to action. LinkedIn Thought Leadership Post:- Hook written strictly within the first 210 characters before the "...see more" fold. 4-point architecture breakdown. Zero-click link prompt (places the link in the first comment). Low-friction debate question to boost comments.