mrbeast-perspective
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSTIME_DELAYED_CONDITIONAL
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external data, such as YouTube subtitles and user-provided video scripts. It possesses capabilities to perform web searches and execute local scripts, creating a surface where malicious instructions embedded in the analyzed content could influence agent behavior.
- Ingestion points: YouTube subtitles (via
scripts/fetch_youtube_subtitles.sh), user scripts (viascripts/retention_curve_checker.py), and video titles (viascripts/analyze_titles.py). - Boundary markers: None identified in the processing workflow or persona instructions.
- Capability inventory: Shell script execution, Python script execution, and web search tool invocation.
- Sanitization: No explicit content sanitization or validation of the processed text is mentioned.
- [COMMAND_EXECUTION]: The skill contains instructions that direct the agent to execute shell commands for maintenance purposes. In the
SKILL.mdfile, the agent is instructed to rungitcommands such asrev-parse HEADandls-remote origin HEADto manage its own update status. - [TIME_DELAYED_CONDITIONAL]: The skill implements a conditional update check mechanism that triggers based on the time elapsed since the last check. The
SKILL.mdfile directs the agent to check a.last-update-checkfile and only perform a version check if it has been more than 30 days since the last execution. - [EXTERNAL_DOWNLOADS]: The provided shell scripts perform automated installation of external packages. The
scripts/fetch_youtube_subtitles.shscript executespip install -q yt-dlpto ensure the well-knownyt-dlputility is available on the system.
Audit Metadata