script-delete

Warn

Audited by Gen Agent Trust Hub on Sep 29, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill recommends the global installation of an external Node.js package, 'unity-mcp-cli', which does not originate from a recognized trusted organization or well-known service. Evidence: The 'Troubleshooting' section explicitly suggests 'npm install -g unity-mcp-cli'.
  • [COMMAND_EXECUTION]: The skill functions by executing shell commands through the 'unity-mcp-cli' tool. Evidence: The 'How to Call' section demonstrates executing the tool via shell commands.
  • [INDIRECT_PROMPT_INJECTION]: The skill takes an array of file paths as input for deletion, creating a vulnerability where an agent influenced by malicious external data (such as a poisoned project manifest) could be coerced into deleting critical system or configuration files. 1. Ingestion points: 'files' parameter in SKILL.md. 2. Boundary markers: None present to restrict deletions to a specific project subdirectory. 3. Capability inventory: File deletion via CLI execution. 4. Sanitization: No sanitization or path validation is described within the skill's instructions.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 29, 2026, 04:32 PM