mrbeast-perspective
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The script
scripts/fetch_youtube_subtitles.shdownloads theyt-dlppackage from a public registry if it is not already present and subsequently fetches data from YouTube servers. - [COMMAND_EXECUTION]: The skill provides and utilizes several shell and Python scripts (
fetch_youtube_subtitles.sh,analyze_titles.py,retention_curve_checker.py,thumbnail_audit.py) to perform local text and image analysis. - [PROMPT_INJECTION]: The skill employs extensive persona-based instructions to simulate the MrBeast character, including strict guidelines for the agent to remain in character and specific "EXIT TRIGGER" and "STOP" markers to manage the roleplay boundaries.
- [INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection by processing external data, such as YouTube subtitles fetched at runtime and user-provided scripts, although its functionality is restricted to content analysis tasks.
- Ingestion points: External data is ingested through
scripts/fetch_youtube_subtitles.sh(YouTube subtitles) andscripts/retention_curve_checker.py(local script files). - Boundary markers: The
SKILL.mdfile defines explicit "EXIT TRIGGER" keywords to return the agent to a standard assistant state. - Capability inventory: The skill has the capability to execute provided shell and Python scripts and perform Web Search operations.
- Sanitization: Shell scripts utilize standard variable quoting to prevent simple injection, and Python scripts use regular expressions for text parsing.
Audit Metadata