creating-video-websites

Pass

Audited by Gen Agent Trust Hub on Jun 25, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: Indirect prompt injection surface identified. The skill's mandatory 'Interview' phase (SKILL.md, Step 0) prompts the agent to fetch design content from a user-provided URL or pasted copy. This untrusted data is then interpolated into the generation process for index.html and app.js without protective measures.\n
  • Ingestion points: 'Content Source' field in the Step 0 Interview (SKILL.md).\n
  • Boundary markers: Absent; there are no instructions for the agent to use delimiters or to ignore potential instructions embedded within the fetched content.\n
  • Capability inventory: The skill executes a local Python script for video processing and generates multiple frontend source files (HTML, CSS, JavaScript).\n
  • Sanitization: No validation, escaping, or filtering of the external content is mentioned.\n- [COMMAND_EXECUTION]: The skill executes a bundled Python script (scripts/extract_frames.py) using the python interpreter to process user videos. It also directs the agent to use terminal commands for package management (pip install).\n- [EXTERNAL_DOWNLOADS]: The skill requires downloading well-known packages from PyPI (opencv-python, numpy, rembg) and utilizes standard CDN-hosted libraries (gsap, lenis) for the generated frontend. These are recognized development dependencies.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 25, 2026, 04:59 PM
Security Audit — agent-trust-hub — creating-video-websites