mcp-skill

Fail

Audited by Gen Agent Trust Hub on Sep 1, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill executes packages from an unknown organization using npx, which downloads and runs code at runtime.\n
  • Evidence in SKILL.md: npx -y @untrusted-org/mcp-data-tool\n
  • Evidence in scripts/setup.py: npx -y @untrusted-org/mcp-processor\n- [EXTERNAL_DOWNLOADS]: The skill initiates network connections to non-whitelisted external domains for MCP server communication.\n
  • Evidence: https://analytics.example.net/sse and https://remote-mcp.example.net/mcp\n- [INDIRECT_PROMPT_INJECTION]: The skill exposes the agent to untrusted data by connecting to external MCP servers without apparent sanitization.\n
  • Ingestion points: External server configurations in SKILL.md and scripts/setup.py.\n
  • Boundary markers: None present to distinguish or protect against malicious tool outputs.\n
  • Capability inventory: Shell command execution via npx and file system writes in setup.py.\n
  • Sanitization: No sanitization or validation of remote server responses identified.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Sep 1, 2026, 11:05 AM
Security Audit — agent-trust-hub — mcp-skill