skills/varghq/skills/find-references/Gen Agent Trust Hub

find-references

Fail

Audited by Gen Agent Trust Hub on Nov 14, 2026

Risk Level: CRITICALINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill aggregates data from over 22 external sources like Are.na and various museum APIs which is then processed by an AI agent to rank references. This ingestion of untrusted external content represents a risk where malicious input could attempt to subvert the agent's logic. * Ingestion points: scripts/search.py and scripts/mcp_call.py. * Boundary markers: Defined prompt structures in judge.py. * Capability inventory: Shell command execution and network fetches. * Sanitization: Limited to HTML stripping in search.py.
  • [COMMAND_EXECUTION]: Multiple Python scripts, including judge.py and frame_light.py, use the subprocess module to run ffmpeg, ffprobe, magick, and AI CLI tools. These operations are essential for the skill's media analysis functions but provide a significant capability surface.
  • [EXTERNAL_DOWNLOADS]: A reference URL in documentation (start-track.com) has been flagged by security scanners as a phishing risk.
  • [DYNAMIC_EXECUTION]: The script motion_read.py conditionally loads media processing libraries cv2 and numpy using dynamic imports. This is restricted to a predefined list of safe packages.
Recommendations
  • CRITICAL: 1 file(s) identified as malware by FileRep - DO NOT USE
  • Contains 2 malicious URL(s) - DO NOT USE
Audit Metadata
Risk Level
CRITICAL
Analyzed
Nov 14, 2026, 09:50 PM