grill-me
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill directs the agent to autonomously fetch information from the filesystem and tools based on a user's plan or decision. This architectural choice enables a scenario where untrusted user input could trigger the agent to read or interact with sensitive local data without explicit user review.
- Ingestion points: Processed user input describing a "plan, decision, or idea" is used to drive exploration.
- Boundary markers: None identified in the instructions to separate user-provided content from agent commands.
- Capability inventory: The skill encourages autonomous filesystem access and tool execution through sub-agents.
- Sanitization: The skill lacks instructions for validating or sanitizing user-provided facts before the agent attempts to verify them in the environment.
- [COMMAND_EXECUTION]: The instructions explicitly tell the agent to use external tools and system resources to find "facts."
- Evidence: "When a frontier question needs a fact from the environment (filesystem, tools, etc.), dispatch a sub-agent to find it; don't ask the user for anything you could look up yourself."
Audit Metadata