| 1 | --- |
| 2 | name: podcast |
| 3 | description: Turn research into a spoken podcast — script, synthesize, stitch. Use when: podcast, generate a podcast, audio briefing, narrate, or turn this into audio. |
| 4 | invocation: model+user |
| 5 | --- |
| 6 | |
| 7 | # Podcast |
| 8 | |
| 9 | ## When to use |
| 10 | Producing a spoken-audio episode from a topic, document set, or briefing — |
| 11 | the full pipeline from research to an mp3. |
| 12 | |
| 13 | ## Setup |
| 14 | Requires `ffmpeg` for stitching (`brew install ffmpeg`, `apt install |
| 15 | ffmpeg`). Voices come from the `tts` skill. Fail loud when `ffmpeg` is missing. |
| 16 | |
| 17 | ## Workflow |
| 18 | 1. Research the topic and draft a spoken script (~150 words per minute of |
| 19 | audio). Read it back for approval before synthesizing. |
| 20 | 2. Synthesize per segment with `tts`; keep segments under 2 minutes each. |
| 21 | 3. Stitch: `ffmpeg -f concat -safe 0 -i segments.txt -codec:a libmp3lame out.mp3`. |
| 22 | 4. Report duration, segment count, and the output path. |
| 23 | |
| 24 | ## Non-goals |
| 25 | - Do not publish or upload the episode anywhere. |
| 26 | - Do not synthesize more than the approved script. |
| 27 | - Do not clone real people's voices (see `tts` non-goals). |
| 28 |