grill
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill follows secure design principles for local development tools. It utilizes a standalone Node.js script to manage the interrogation process with built-in security features like path traversal checks and Content Security Policy (CSP).\n- [COMMAND_EXECUTION]: The skill executes its own helper script
scripts/decision_form.mjsto serve a web-based decision form. This script is strictly local, restricting connections to127.0.0.1and requiring a cryptographically random token for access.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes project-specific data and user feedback which could theoretically contain injection attempts.\n - Ingestion points: Ingests data from investigation briefs, local source code files, and user-provided answers from the decision form.\n
- Boundary markers: Uses structured JSON for question configuration and answer collection, separating control logic from user data.\n
- Capability inventory: Possesses file system read/write access for decision records and the ability to execute its local Node.js helper script.\n
- Sanitization: The helper script implements HTML escaping for the interrogation UI and strictly validates JSON responses before they are processed by the agent.
Audit Metadata