product-reel-generator

Pass

Audited by Gen Agent Trust Hub on Jun 23, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: Indirect Prompt Injection Surface (Category 8). The skill extracts product information (names, descriptions, prices) from user-provided e-commerce URLs and interpolates this data into AI video generation prompts and FFmpeg text overlays.
  • Ingestion points: External data is ingested during Step 1 (image/metadata scraping) in SKILL.md and via the --image-url and --prompt arguments in scripts/higgsfield_video.py.
  • Boundary markers: No delimiters or instructions are provided to the agent to treat scraped content as untrusted or to ignore embedded instructions.
  • Capability inventory: The skill possesses extensive capabilities including file system access (Write, Edit), network access, and the ability to execute shell commands (Bash, FFmpeg).
  • Sanitization: There is no evidence of sanitization or validation of the scraped strings before they are used in sensitive generation contexts.
  • [EXTERNAL_DOWNLOADS]: The skill downloads image assets from arbitrary e-commerce domains and fetches generated video files from the Higgsfield AI platform (platform.higgsfield.ai).
  • [COMMAND_EXECUTION]: Utilizes the Bash tool to execute complex FFmpeg command lines for video stitching, Ken Burns effect generation, and audio mixing. It also executes a local Python script (scripts/higgsfield_video.py) to manage API interactions.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 23, 2026, 02:14 PM
Security Audit — agent-trust-hub — product-reel-generator