grilling
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill relies on user-provided answers to build and navigate a decision tree. Malicious content within these answers could potentially influence the agent's behavior in subsequent rounds, particularly when it determines which 'facts' to investigate.
- Ingestion points: User responses to frontier questions in SKILL.md.
- Boundary markers: The instructions do not define specific delimiters for separating user input from system instructions.
- Capability inventory: Mentions access to filesystem and system tools via sub-agents in SKILL.md.
- Sanitization: No explicit sanitization or validation of user answers is described.
- [COMMAND_EXECUTION]: The skill explicitly directs the agent to perform autonomous environment exploration ('Finding facts is your job... dispatch a sub-agent to find it') using the filesystem and available tools. While this is part of the intended functionality, it grants the agent an active role in executing system-level operations based on the conversational context.
Audit Metadata