podcast-transcript-fetcher
Pass
Audited by Gen Agent Trust Hub on Jul 2, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/get_transcript.pyexecutes theffmpegcommand viasubprocess.runto perform audio compression. This operation is used to ensure audio files stay within the 25MB limit required by the Groq Whisper API. The command is invoked using a structured argument list without a shell, minimizing injection risks. - [EXTERNAL_DOWNLOADS]: The skill fetches data from several external domains to obtain podcast metadata, audio files, and transcripts. Sources include RSS feeds from Substack, Libsyn, Transistor.fm, and Simplecast, as well as transcript archives on GitHub Gists and Dropbox. It also interacts with official APIs from Groq and Taddy.
- [PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection (Category 8) because it ingests untrusted external content (podcast transcripts and RSS metadata) which is then passed to the AI agent for analysis.
- Ingestion points: Podcast transcripts and metadata are fetched from multiple external web sources in
scripts/get_transcript.py. - Boundary markers: The script does not use explicit delimiters or "ignore instructions" warnings when providing the fetched content to the agent.
- Capability inventory: The script possesses network access capabilities and subprocess execution (ffmpeg).
- Sanitization: The script performs formatting cleanup (HTML unescaping and regex cleaning) but does not validate the content for malicious agent instructions.
Audit Metadata