youtube-outliers
Pass
Audited by Gen Agent Trust Hub on Mar 27, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes 'yt-dlp' via the subprocess module in
scripts/scrape_youtube_outliers.py. This is a legitimate use of the tool to fetch YouTube metadata (views, upload dates, channel info) without downloading video content. The arguments are passed as a list, and input (keywords) is interpolated into a search string, which follows standard usage patterns for this utility. - [DATA_EXPOSURE_AND_EXFILTRATION]: The skill requires several API tokens (APIFY_API_TOKEN, ANTHROPIC_API_KEY) and Google service account credentials. The instructions correctly advise storing these in environment variables or a
.envfile, which is a standard and safe practice for local agent tools. - [REMOTE_CODE_EXECUTION]: The skill uses the Apify platform to run a remote 'actor' (
karamelo/youtube-transcripts) to fetch video transcripts. This is a controlled execution of a specific, named remote service provided by a third-party platform and does not constitute arbitrary remote code execution. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data (YouTube transcripts) by sending them to an LLM (Claude) for summarization in
summarize_transcript. While this presents a surface for indirect prompt injection, the risk is mitigated by the specific summarization prompt and the low-privilege nature of the summary output (stored in a spreadsheet).
Audit Metadata