generate_podcast

Installation
SKILL.md

Generate Podcast

Purpose

Generate podcasts, audio briefings, and narrated summaries. For voice messages, artifact audio, or reading supplied text aloud verbatim, use tts. The podcast-helper script handles synthesis, catalog management, cover art generation, and publishing.

Workflow

1. Plan

  • Topic and angle
  • Speaker names and voices — prefer the Meta AI voices by default (the MAI_01 and MAI_03 catalog ids: Warm and Smooth); they are the recommended, production-quality set and the right pick most of the time. Default to Warm (avocado_v2:MAI_01) and Smooth (avocado_v2:MAI_03). When the topic, tone, or user request steers toward a different catalog voice — an accent, a character fit, or more distinct speakers — it's perfectly fine to pick it from the catalog. The full voice catalog is at /opt/hatch/skills/voice-selector/voice_source.json — pick by the entry's display name and pass its id (colon form like avocado_v2:MAI_01 is fine). Any number of speakers is supported. ("MAI" is the internal id prefix only — say "Meta AI voices" or display names to the user, never "MAI".)
  • Host names — give each speaker a real first name (e.g. Alex, Jordan). These are the show's own character names, chosen by you and separate from the voice display names: never label a speaker with a voice name like "Warm" or "Smooth". Remember the name you give each voice — record the pairing in ~/MEMORY.md (e.g. "podcast host Alex = avocado_v2:MAI_01") and, before naming a new host, check there for a name you've already used for that voice. Reuse the same name whenever you use that voice again, so a given voice is always the same character to the user. For a recurring series, reuse the same host names and the same voice assignments in every episode so it sounds like the same show — pin them in the cron task body (see Scheduling).
  • Target length — 5–15 minutes; ~130 words/minute
  • Avoid repeating past episodes — before settling the topic and angle, run podcast-helper manifest read and look at each recent episode's topics field (a short summary of what that episode already covered). Steer the new episode toward fresh material or a genuinely new angle rather than re-covering ground. This matters most for a recurring series.
Installs
2
GitHub Stars
20
First Seen
Oct 6, 2026