eeailab-montessori

Pass

Audited by Gen Agent Trust Hub on Nov 13, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The script scripts/video_to_observation_pack.py utilizes the subprocess module to execute system binaries ffmpeg and ffprobe. These calls are used to extract keyframes, audio, and metadata from local video files for behavioral analysis. The implementation correctly uses argument lists rather than shell strings, which mitigates common injection risks.
  • [INDIRECT_PROMPT_INJECTION]: The video processing workflow generates a Markdown file (observation_pack.md) that combines user-supplied strings with instructions for the AI agent. This creates a potential surface where untrusted data could influence subsequent agent reasoning. 1. Ingestion points: User-provided text strings (e.g., child's age and specific parent questions) are ingested as command-line arguments to the script and written directly to the output file. 2. Boundary markers: The generated file uses Markdown headers and lists to separate user data from analysis hints, providing basic delimitation. 3. Capability inventory: The skill possesses file-writing capabilities and subprocess execution for media processing tasks. 4. Sanitization: User inputs are interpolated into the Markdown template without explicit escaping or sanitization, allowing for potential structural modification of the file content.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 13, 2026, 12:51 PM