🛡️ Structured Output & Function Calling Guard

When AI agents output structured function call arguments, attackers can attempt: 1. JSON Parameter Smuggling: Embedding prompt injection directives inside JSON string fields. 2. Path Traversal: Injecting ../ or /etc/passwd into file path parameters. 3. Command Chaining: Injecting ;, &&, or backticks into shell execution arguments.

StructuredOutputGuard validates structured arguments before tool execution.

---

🚀 Quick Example

from lightguard import StructuredOutputGuard

guard = StructuredOutputGuard()

validation = guard.validate_tool_arguments(
    tool_name="read_file",
    arguments={"path": "../../etc/shadow"}
)

if not validation["allowed"]:
    print("Action Blocked:", validation["violations"])