ai-infrastructure-modal

Fail

Audited by Gen Agent Trust Hub on Apr 7, 2026

Risk Level: CRITICALREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The reference.md file provides an image-building example that uses a highly dangerous pattern (curl | bash). This facilitates unverified remote script execution and is a severe security risk if implemented with untrusted URLs.
  • [COMMAND_EXECUTION]: The examples/core.md file demonstrates the usage of subprocess.Popen to launch external server processes like vLLM, which creates a significant command execution surface.
  • [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection. 1. Ingestion points: SKILL.md (predict function), examples/core.md (classify and generate functions). 2. Boundary markers: Absent. 3. Capability inventory: subprocess.Popen in examples/core.md and run_commands in reference.md. 4. Sanitization: Absent. The skill accepts arbitrary dictionary payloads via FastAPI endpoints and passes them to underlying GPU functions without validation.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
CRITICAL
Analyzed
Apr 7, 2026, 01:31 AM