ai-marketing-videos

Fail

Audited by Gen Agent Trust Hub on Apr 25, 2026

Risk Level: CRITICALREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill uses 'curl -fsSL https://cli.inference.sh | sh' to install its CLI tool. This is a critical security risk as it executes code from a remote server directly in the shell without any verification of content or integrity.
  • [EXTERNAL_DOWNLOADS]: The skill downloads an installation script from a remote URL that is not recognized as a trusted or well-known service.
  • [COMMAND_EXECUTION]: The skill invokes numerous shell commands via the 'infsh' utility to manage video generation, model execution, and authentication.
  • [CREDENTIALS_UNSAFE]: The 'infsh login' command indicates the skill handles authentication credentials, which could be harvested if the environment is compromised.
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting data from generated files (e.g., script.json) and passing it directly to AI model prompts.
  • Ingestion points: Reads from script.json, voiceover.json, and multiple section-specific JSON files.
  • Boundary markers: Absent; the skill does not use delimiters to isolate untrusted data from the system prompt.
  • Capability inventory: Executes external AI models and media processing tools via the infsh command.
  • Sanitization: No evidence of escaping or validation of data retrieved from external or generated sources before interpolation into shell commands.
Recommendations
  • HIGH: Downloads and executes remote code from: https://cli.inference.sh - DO NOT USE without thorough review
  • AI detected serious security threats
Audit Metadata
Risk Level
CRITICAL
Analyzed
Apr 25, 2026, 06:52 PM
Security Audit — agent-trust-hub — ai-marketing-videos