grill-me-qa
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates an attack surface by exploring untrusted user data at runtime.
- Ingestion points: In Phase 1, the agent is directed to "Explore codebase, test suite, CI config, existing docs" (SKILL.md) to gather context for the grilling session.
- Boundary markers: The instructions do not define clear boundaries or provide explicit "ignore embedded instructions" warnings for the content read from the user's codebase.
- Capability inventory: The skill's primary function is information gathering and recommendation generation (TSDR). It does not invoke high-privilege tools like arbitrary command execution or external network requests directly within its own instructions.
- Sanitization: There are no mentioned mechanisms for sanitizing, escaping, or validating the contents of the files explored during the context discovery phase before they are used to influence agent decisions.
Audit Metadata