How to use the AI Podcast Voiceover agent skill

What the AI Podcast Voiceover skill makes the agent do

An agent skill is a folder of instructions and supporting files that an AI agent reads when it is relevant to a task. This one covers podcast voiceover work: it takes written material and produces a solo episode with a consistent host voice, natural pacing, pronunciation control, and an MP3 file you can publish.

The synthesis itself runs on the hosted, paid Beatra service at mcp.beatra.ai. The catalog entry is a reviewed pointer to an external package. It contains no client code and performs no Beatra operation by itself.

• Works from an article, notes, or a finished script
• Produces a solo episode with a consistent host voice
• Returns an MP3 you can publish
• Runs inside Claude Code, Codex, or OpenClaw, according to the package description

What you need before you install

Check these requirements first. The package installs Python scripts and opens a browser sign-in, so you need to be comfortable reviewing code and granting access.

• Python 3.10 or newer
• git, used to fetch the pinned package
• A Beatra account with paid credits, plus browser sign-in to authorize the skill
• Network access to github.com and mcp.beatra.ai

Install the skill from a pinned, reviewed copy

The package documents a git-based install that fetches one exact commit and checks its tree ID before anything is copied. Run the fetch only after you approve it, because it downloads code from github.com. The directory page also offers a ZIP download: you can unzip it into ~/.claude/skills/ for your user, or into a project's .claude/skills/ folder. In the Claude apps, you can upload the skill under Settings → Capabilities → Skills instead.

Stop if the last line does not print TREE-OK. The expected IDs come from this catalog entry, not from the package or its website.

Review the files and copy them in

Inspect the tree before activation. The first command lists any symlink, executable, or submodule entry, and the second searches for the entry points of an older self-updater. Both are inventory aids only. You must read every file yourself, including each script, reference, and hook, before you treat the package as safe. If you cannot do that, stop and ask for a separate review.

The package contains SKILL.md, manifest.json, install-source.json, bundled Markdown references, and three Python scripts: scripts/authorize.py, scripts/mcp_client.py, and scripts/uninstall.py. Copying into your skills folder changes agent configuration, so approve that as a separate decision.

• Confirm the tree holds regular files only (no symlinks, executable bits, or binaries)
• Check that the destination folder does not already exist before copying
• Copy the reviewed folder into ~/.claude/skills/ai-podcast-voiceover

Authorize the skill and run a free connection check

Authorization opens a browser sign-in and stores a bearer token on your machine. Only run it when you are ready to connect the skill to your Beatra account. Start a new agent session afterwards if your host only discovers skills at startup.

The connection check is non-billable, so it is a safe first test that the skill is installed and reachable.

• Authorization stores a bearer token, so decide to run it deliberately
• The verify command does not create billable work

Ask for a podcast episode in plain language

Once the skill is installed and authorized, describe the episode you want and point the agent at your material. The skill reads its own instructions, then uses free discovery calls to check current model cards and credit estimates before it proposes any paid step.

Expect a cost card before any paid call. It shows the tool, the model, the estimate, and exactly which files will be uploaded. Planning or a phrase like just do it does not count as approval, so answer the cost card explicitly.

• Turn this article into a five-minute solo episode with my host voice, and show me the cost before anything is synthesized.
• Use the notes in this folder to draft and voice a short episode. Upload only the script file.
• Voice this finished script for my podcast and give me the MP3 path once it is delivered.
• Check the current model options and credit estimates for a solo episode from this draft, but do not submit anything yet.

How a paid job runs, step by step

The skill follows a fixed loop. It reads the installed instructions, runs free discovery, shows one cost card, uploads only files you supplied for this task, and submits once with a stable request identifier. It then polls the returned task.

After delivery, the agent reports the artifact and the net charge in billing.net_charged_credits. It asks again before any further paid stage, so each paid step needs its own approval.

• Discovery is free and non-billable
• One cost card per paid call, with explicit approval
• Only files you approved are uploaded
• If a response is lost, retry only with the same request ID and identical arguments

Security, updates and uninstalling

The skill sends scripts and any uploaded files to a third party for hosted synthesis. It also stores a bearer token and uses network access to github.com and mcp.beatra.ai. Do not use it for confidential or unpublished material you do not want uploaded, and do not use it for local text-to-speech, which the skill says not to do.

This version cannot update itself. To move to a newer version, repeat the review steps for that commit in a new review directory, compare the file-level changes, and replace the folder only after approval. Versions not listed in the catalog entry have not been reviewed.

To uninstall, run the dry run first. Uninstall reports disconnected when this was the last Beatra skill on the device, which revokes the token and removes local Beatra credential and state files. keep_connection means another Beatra skill still uses the shared token, and revoke_retry means the server refused revocation, so run it again shortly.

• Your script and uploaded files leave your machine for synthesis
• The install adds executable Python code and stores a bearer token
• The package has no self-updater that downloads code
• Read the JSON decision from the uninstall dry run before running it for real