How to use the Podcast Generation agent skill

What does the podcast-generation skill make the agent do?

Once the skill is installed, the agent can take text you provide and rewrite it as a casual conversation between a male and a female host. The skill describes a workflow: the agent first writes a structured JSON script of dialogue, then runs a bundled Python script that sends the lines to a text-to-speech service and assembles the audio into one MP3 file.

The skill targets roughly 10 minutes of dialogue, usually around 40 to 60 lines, and supports English and Chinese content. Each episode is meant to open with the male host greeting listeners with "Hello Deer".

• Converts articles, reports and documentation into podcast scripts
• Produces two-host dialogue with natural back-and-forth
• Synthesizes speech and mixes the chunks into a podcast MP3
• Saves a readable markdown transcript alongside the audio
• Supports English (en) and Chinese (zh) content

How do you install the podcast-generation skill?

Download the ZIP file for the skill from this page. You can install it in one of two ways, depending on how you use your agent.

For Claude Code style setups, unzip the download into your personal skills folder ~/.claude/skills/ or into a project-level .claude/skills/ folder inside your project. For the Claude apps, upload the ZIP under Settings → Capabilities → Skills.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into .claude/skills/ in your project
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

What files come with the skill?

The download contains three files that the agent uses. SKILL.md holds the instructions the agent reads, scripts/generate.py performs the audio generation, and templates/tech-explainer.md is a template for converting technical documentation and tutorials into audio-friendly dialogue.

• SKILL.md: the instructions and workflow
• scripts/generate.py: the script that calls the text-to-speech service and builds the MP3
• templates/tech-explainer.md: guidance for turning technical content into a podcast

How do you set up the speech service credentials?

The script cannot produce audio without a text-to-speech provider. The source supports two providers, MiniMax and Volcengine, and selects them through environment variables. You need an account and API key with one of them.

For Volcengine, set VOLCENGINE_TTS_APPID and VOLCENGINE_TTS_ACCESS_TOKEN. For MiniMax, set MINIMAX_API_KEY. You can optionally set VOLCENGINE_TTS_CLUSTER, VOLCENGINE_TTS_VOICE_TYPE_MALE and VOLCENGINE_TTS_VOICE_TYPE_FEMALE. If these are unset or blank, the defaults listed in the skill are used.

• Volcengine: VOLCENGINE_TTS_APPID and VOLCENGINE_TTS_ACCESS_TOKEN
• MiniMax: MINIMAX_API_KEY
• Optional Volcengine settings: VOLCENGINE_TTS_CLUSTER and the male and female voice type variables
• Python with the requests library must be available

How do you trigger the skill with a natural request?

You do not need special syntax. Ask for a podcast in plain language, and provide the content or point the agent to the file. The skill's description tells the agent to use it when you ask to generate, create or produce podcasts from text.

Good requests name the source material and the format you want. The agent identifies the source content, the language and where to save the output before it starts.

• "Generate a podcast about the history of artificial intelligence"
• "Turn this article into a podcast episode"
• "Create an audio briefing from this internal documentation"
• "Make a podcast from this report in Chinese"

What output do you get and how is it checked?

The skill saves the MP3 and transcript to /mnt/user-data/outputs/. The agent is instructed to share both files and give a brief summary of the topic, duration and hosts, then offer to regenerate if you want changes.

The skill states that if the script exits with status 1, generation failed and diagnostics are written to stderr. An existing audio file or a new transcript alone does not prove that audio was generated successfully, so check that the run succeeded before relying on the MP3.

• Podcast MP3 saved to /mnt/user-data/outputs/
• Markdown transcript saved alongside it
• Failure is signalled by exit status 1 and stderr messages

What should you check before using it?

The script sends the text to third-party speech APIs over the network, so do not use confidential documents. The agent also rewrites material into dialogue and may simplify or add errors, so content that must be exact or legally verified needs checking against the original.

The skill also expects access to the /mnt/user-data and /mnt/skills file paths used by the agent runtime, and it runs a script. Review the files before installing if you need to control what runs on your machine.

• Network: text is sent to the chosen TTS provider
• Scripts: the skill runs a bundled Python script
• Accuracy: the rewritten dialogue can simplify or introduce errors
• Privacy: avoid confidential material

What tips help get a better podcast?

Give the agent clean source text. The skill's guidelines say technical content should be simplified, code and formulas replaced with plain descriptions, and metadata such as dates and author names left out. The template for technical documentation recommends describing code in words rather than pasting it.

Keep your request focused on one topic. If the result needs changes, ask the agent to regenerate it with your adjustments.

• Paste or point to a single, well-structured article or report
• For technical docs, ask for plain-language explanations instead of code
• Ask for Chinese output explicitly if your content is in Chinese
• Ask for a regeneration if the tone or facts need correcting