Podcastfy (Clawdbot)

Generate a podcast-style audio conversation (MP3) from a URL (or multiple URLs) using podcastfy.

This skill provides a wrapper script that:

• creates/uses a local venv ({baseDir}/.venv)
• installs/updates podcastfy
• runs Podcastfy with Gemini for transcript generation and Edge for TTS

One-time setup

1) Ensure ffmpeg is installed on the host.

Ubuntu/Debian:

sudo apt-get update && sudo apt-get install -y ffmpeg

2) Provide a Gemini API key:

Set env var GEMINI_API_KEY (recommended), or create a project .env and export it before running.

Example keys file: {baseDir}/references/env.example

Quick start

Generate an MP3 from a single URL:

cd {baseDir}
export GEMINI_API_KEY="..."
./scripts/podcastfy_generate.py --url "https://example.com/article"

Multiple URLs:

cd {baseDir}
./scripts/podcastfy_generate.py --url "https://a" --url "https://b"

Long-form:

cd {baseDir}
./scripts/podcastfy_generate.py --url "https://example.com/long" --longform

Output

The script writes outputs under:

• {baseDir}/output/audio/ (MP3)
• {baseDir}/output/transcripts/ (transcript)

Podcastfy prints the final MP3 path on success.

Optional tuning

• PODCASTFY_LLM_MODEL (default: gemini-1.5-flash)
• PODCASTFY_EDGE_VOICE_Q (default: en-US-JennyNeural)
• PODCASTFY_EDGE_VOICE_A (default: en-US-EricNeural)

Automation / reliability tips (important)

Prefer RSS feeds over browser automation for cron jobs

In isolated cron jobs, avoid relying on a system browser (Chrome/Chromium) or extra Python deps (e.g. bs4).

For TechCrunch categories, prefer the RSS feed:

• https://techcrunch.com/category/artificial-intelligence/feed/

This reduces breakage from:

• “No supported browser found …”
• missing site parsing dependencies in the cron runtime

Validate MP3 output (avoid 0-second audio)

We observed podcastfy can occasionally produce an MP3 that is effectively empty/truncated (e.g., a ~261-byte file), which shows up as 0s in Telegram.

This wrapper now validates MP3 output and will automatically fall back to edge-tts synthesis from the latest transcript when the MP3 is invalid.

Safety / workflow

• Prefer “draft-first”: tell the user what will be generated (language/length) before running.
• Never paste API keys into chat logs.