GuardrailMiddleware

Block or reject tool invocations by tool name and argument regex policies

Class: bxModules.bxai.models.middleware.core.GuardrailMiddleware

Guards tool execution only. It can block specific tools or reject calls whose arguments match forbidden regex patterns.

Features

• Denylist tool names with blockedTools
• Validate arguments per tool with argPatterns
• Works with both normalized toolArgs and provider-native tool call shapes
• Returns reject() with reason when a rule matches

Constructor

middleware = new bxModules.bxai.models.middleware.core.GuardrailMiddleware(
    blockedTools: [ "deleteRecord", "dropTable" ],
    argPatterns : {
        runSQL: [ { query: "^SELECT" } ]
    }
)

Configuration

Option	Type	Default	Description
blockedTools	array	[]	Tool names that are always rejected
argPatterns	struct	{}	Per-tool regex rules in the form { toolName: [ { paramName: "pattern" } ] }

Hooks Used

• beforeToolCall

Example

agent = aiAgent(
    tools      : [ runSQLTool, deleteUserTool ],
    middleware : [
        new bxModules.bxai.models.middleware.core.GuardrailMiddleware(
            blockedTools: [ "deleteUser" ],
            argPatterns : {
                runSQL: [
                    { query: "(?i)^\\s*select\\b" },
                    { query: "(?i)drop|truncate|delete" }
                ]
            }
        )
    ]
)

Notes

• Tool-name matching for blockedTools is case-insensitive.
• Pattern rules run only for tools present in argPatterns.
• This middleware does not sanitize prompt text or model output.