animated-explainer-ad

Warn

Audited by Gen Agent Trust Hub on Jul 20, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The script scripts/auto.sh is vulnerable to command injection via shell variable interpolation. It passes a JSON object ($META) into a python3 -c command using triple-quotes ('''$META'''). If a user-provided value (such as a character descriptor) contains the sequence ''', an attacker can break out of the Python string literal and execute arbitrary Python code. This risk is present when running the skill in 'autopilot' mode.
  • [COMMAND_EXECUTION]: The skill makes extensive use of the subprocess module across several scripts (render_clip.py, build_endcard.py, render_vo.py, lib.py) to execute shell commands for media processing with ffmpeg and ffprobe. While these specific calls use argument lists and appear safe from direct shell injection, they represent a significant execution capability within the skill's environment.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its architecture of ingesting free-form user data to generate a brief.md file, which then drives the entire asset generation pipeline.
  • Ingestion points: User input gathered during Phase 1 (Intake) is used to define brand details, product claims, and character descriptions in brief.md (e.g., in SKILL.md Phase 1 instructions).
  • Boundary markers: The skill does not implement boundary markers or instructions to ignore embedded commands when processing the content of brief.md or constructing prompts for external AI models.
  • Capability inventory: The skill can execute arbitrary shell commands via scripts, make network requests to external APIs, and read/write to the local filesystem.
  • Sanitization: No sanitization or validation is performed on the content extracted from user prompts before it is used to build command-line arguments or interpolated into the vulnerable auto.sh script.
  • [EXTERNAL_DOWNLOADS]: The skill downloads generated media assets from well-known services, including FAL (fal.ai) for images and video, and ElevenLabs (elevenlabs.io) for audio and music, which are functional requirements for its stated purpose.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 20, 2026, 06:25 PM
Security Audit — agent-trust-hub — animated-explainer-ad