world-voice

Pass

Audited by Gen Agent Trust Hub on Mar 6, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform basic file system management operations.\n
  • Evidence: Execution of mkdir -p world-code to prepare a directory for output and managing the voice.md file within that directory.\n- [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection due to the processing of external user-provided data.\n
  • Ingestion points: Step 3, Question 1 explicitly asks the user to provide an external writing sample (e.g., email or social post) for analysis.\n
  • Boundary markers: There are no explicit markers or instructions defined to prevent the agent from executing instructions that might be embedded in the user's writing sample.\n
  • Capability inventory: The skill has access to bash tools for directory creation and file read/write operations.\n
  • Sanitization: The instructions do not include any steps for sanitizing, escaping, or validating the user-provided text before analysis.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 6, 2026, 09:28 PM