aip-reference-style
Warn
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: MEDIUMPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill instructs the agent to install system dependencies such as ffmpeg, python3, and uv without user approval, bypassing interactive safety confirmations for environment changes.\n- [COMMAND_EXECUTION]: The skill relies on extensive use of
subprocess.runto execute ffmpeg and ffprobe on user-supplied reference media files.\n- [EXTERNAL_DOWNLOADS]: The agent is directed to download media from social platforms (Instagram, TikTok, YouTube) using backend-provided URLs and fetch font-related resources.\n- [REMOTE_CODE_EXECUTION]: The skill usesuv run --with fonttoolsto dynamically fetch and execute the fonttools library from public registries during font specimen generation.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted media from external links, creating an ingestion surface.\n - Ingestion points: Reference videos and images from user-supplied social media URLs.\n
- Boundary markers: None present.\n
- Capability inventory: Shell command execution, file system modification, and system tool installation.\n
- Sanitization: Uses list-based arguments in
subprocess.runto mitigate shell injection risks.
Audit Metadata