Generate a podcast episode from content you provide. Paste text, point to local files, or describe a topic. Two AI hosts discuss it in a natural conversation. Listen locally or in your favorite podcast app via RSS.
Throughout this file:
SKILL_DIR = the directory where this SKILL.md file lives (the skill's install location)DATA_DIR = ~/.personalized-podcast (where user data, config, and episodes are stored)Resolve these to absolute paths before running any commands.
When the user invokes this skill, start by introducing it:
"This skill turns any content into a podcast episode with two AI hosts who discuss it in a natural, conversational style. Think NotebookLM, but you control everything: the script prompt, the hosts' voices, the show format.
You can use it to:
- Listen to anything on the go. Paste an article, point to a file, drop a URL. Instead of reading, you get two people breaking it down while you walk, commute, or cook.
- Understand yourself from the outside. Feed it personal content like meeting transcripts, your resume, or journal entries, and have the hosts share their impressions of you. It's genuinely illuminating.
Let's get started. What content do you want to turn into a podcast?"
If the user already provided content with their /podcast command, skip the question and go straight to setup check.
If DATA_DIR/config.yaml does not exist, run through this setup BEFORE generating. Do each step yourself - don't ask the user to run commands.
mkdir -p DATA_DIR/{scripts_output,episodes,logs}
python3 -m venv DATA_DIR/venv
DATA_DIR/venv/bin/pip install httpx pydub pyyaml python-dotenv jinja2 audioop-lts
Check ffmpeg is installed:
ffmpeg -version
If not found, install it: brew install ffmpeg (macOS) or sudo apt install ffmpeg (Linux).
cp SKILL_DIR/config/config.example.yaml DATA_DIR/config.yaml
The config comes with two pre-picked voices that work out of the box. No voice selection needed for first use.
Tell the user:
"To turn your script into audio, this skill uses Fish Audio for text-to-speech. They have a free tier with 2 million+ voices. You just need an API key:
- Go to https://fish.audio and create a free account
- Go to https://fish.audio/app/api-keys/ and create an API key"
Then create the .env file and open it:
echo "FISH_API_KEY=your_key_here" > DATA_DIR/.env
open DATA_DIR/.env
Tell them: "Paste your Fish Audio API key in this file, replacing 'your_key_here'. Save and close."
IMPORTANT: Never ask the user to paste API keys in the chat. Always use the .env file.
Tell the user:
"You're all set! The default show has two hosts - Alex (curious, energetic) and Sam (analytical, witty) - with voices pre-configured. Let's generate your first episode."
Then proceed to generate the episode. Do NOT set up RSS or GitHub Pages during first-time setup. That's optional and comes later.
Read all the content the user provided. If they pointed to files, read them with the Read tool. If they pasted a URL, fetch it. Combine everything into your understanding of the source material.
Read the prompt file at SKILL_DIR/PROMPT.md for the hosts, style, structure, and output format. Follow those instructions to write the script.
If the user included custom instructions in their /podcast message (e.g., "make it a debate" or "hosts should eavesdrop on my conversation"), incorporate those. The user's inline instructions override PROMPT.md for that episode.
Save the script as a JSON array using the Write tool to: DATA_DIR/scripts_output/YYYY-MM-DD.json (use today's date). If a file for today already exists, append a number (e.g., 2026-04-05-2.json).
Tell the user: "Script written. Now generating audio - this takes about a minute depending on episode length."
Run the speak script:
DATA_DIR/venv/bin/python SKILL_DIR/scripts/speak.py --script <path_to_script.json>
This outputs the path to the generated MP3 file.
IMMEDIATELY open the audio file for the user:
open <path_to_mp3>
On Linux use xdg-open instead of open.
After the audio plays, tell the user:
"Your episode is ready! A few things you can do from here:
Try different show formats. The default is two hosts chatting, but you can do anything. Just describe it when you run
/podcast:
- "/podcast make it a debate about this article"
- "/podcast hosts should eavesdrop on my conversation and share their impressions of me"
- "/podcast interview format, one host asks questions and the other is the expert"
- "/podcast solo narrator, walk me through this research paper"
- "/podcast news roundup, read each tweet aloud then discuss"
Or edit
PROMPT.mdto permanently change your show's format.Pick your own voices. The default voices are solid, but Fish Audio has 2 million+ to choose from. Browse https://fish.audio/discovery, find two you like, and update the voice IDs in
DATA_DIR/config.yaml.Listen in your podcast app. Want new episodes delivered to Apple Podcasts, Spotify, Overcast, or Snipd automatically? Just ask me to set up an RSS feed for you - takes about a minute."
Only show these tips the FIRST time, or if the user asks about customization.
When the user wants to set up an RSS feed, explain why it's useful:
"This creates a personal podcast feed that works in any podcast app. Once it's set up, every time you generate a new episode, it gets published to the feed and shows up where you already listen to podcasts. No new app needed."
Then walk them through these steps:
Ask for their GitHub username, then:
gh repo create USERNAME/podcast-feed --public --description "Personal podcast feed"
cd /tmp && mkdir podcast-feed && cd podcast-feed && git init
echo '[]' > episodes.json
Generate an initial empty feed.xml using the template, then push and enable GitHub Pages:
gh api repos/USERNAME/podcast-feed/pages -X POST --input - << 'EOF'
{"source":{"branch":"main","path":"/"},"build_type":"legacy"}
EOF
Update DATA_DIR/config.yaml with: