The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests untrusted content (YouTube titles and transcripts) and interpolates it directly into prompts for summarization and title generation.
Ingestion points: scripts/scrape_cross_niche_outliers.py and scripts/scrape_cross_niche_tubelab.py fetch transcripts and titles from YouTube via external APIs and scraping tools.
Boundary markers: Absent. The transcripts are placed directly into LLM prompts (e.g., Transcript (first 8000 chars): {text[:8000]}) without delimiters or instructions to ignore embedded commands.
Capability inventory: The skill has the ability to execute shell commands (yt-dlp) and perform authenticated writes to Google Sheets.
Sanitization: No sanitization or filtering of the external transcript text is performed before it is processed by the model.
[COMMAND_EXECUTION]: The script scripts/scrape_cross_niche_outliers.py uses subprocess.run to execute the yt-dlp command-line utility. While the script correctly passes arguments as a list to mitigate shell injection, the execution of external binaries is a high-privilege operation that depends on the environment having the tool correctly installed.

cross-niche-outliers