Edit and publish captioned YouTube Shorts with Codex and FFmpeg

I used Codex and FFmpeg to turn two How I AI recordings into eight vertical Shorts, then had Codex handle the repetitive publishing work in YouTube Studio. The biggest improvement came from giving it an actual editing brief. “Make it punchy” did not explain when I wanted my face, a split screen, or the demo.

1. Start with the recordings and candidate moments. Give the agent the section you want to use, then have it find a complete opening and ending. A funny line is not a useful clip if it starts halfway through the setup.
2. Set the shot rules. I wanted full-screen reactions for funny or dramatic lines, a steady 50/50 view while explaining something, and screen capture when the screen actually mattered. The cuts needed to follow what I was saying.
3. Review the rendered clips. Codex made the cuts, framing, and captions with local video tools. I watched the exports and asked for revisions to the framing and clipped thoughts. Watching the video caught things a plan or a transcript could not.
4. Delegate the upload work after approval. Codex operated YouTube Studio in the browser to upload the finished clips, add titles and descriptions, and connect each Short to the relevant full episode.

Here is Claude Is Not a Party Boy, one of the finished Shorts. I still made the editorial calls. Having the agent execute the revisions and work through the upload screens made it much easier to get the whole set out.

# Edit this recording into a Short

Source recording: [file or accessible link]
Moment to use: [time range or description]
Target length and format: [for example, a vertical 9:16 Short]

Use only the footage I supply. First identify a complete opening thought and a natural ending. Preserve enough setup for someone who has not seen the episode.

Choose framing based on the dialogue:
- Full-screen speaker for an introduction, punchline, or reaction.
- A fixed 50/50 speaker-and-screen view for explanations that need both.
- Full-screen screen capture when the viewer needs to see the detail.
- A close crop of the relevant screen area only when it remains readable.
- Return to the speaker when the screen stops adding information.

Do not make constant cuts just to create motion. Hold a useful frame. Keep the split-screen layout fixed through a continuous explanation.

Add readable, speech-timed captions that avoid faces and important screen content. Listen through the opening and ending so neither cuts off a word or leaves an unfinished thought.

Render an actual review file. After my feedback, revise the file and verify its audio, captions, framing, and ending. Prepare a title, description, and related full-episode link. Upload or publish only after I approve the final clip and destination.