Recipes

Guard email

For an agent that sends email, in 15 frameworks.

Last reviewed 2026-10-02

Mail inside your company runs; mail to anyone outside waits for a person. Recipients are also destinations: until you allow a destination under egress rules, the egress stage refuses it on its own.

The rule

{
  "name": "email-guard",
  "effect": "require_approval",
  "priority": 10,
  "match_path": "/tools/send_email",
  "conditions": [
    {
      "field": "to",
      "op": "not_contains",
      "value": "@yourcompany.com"
    }
  ]
}

Measured on production (2026-10-02, rule tester): the risky call below matches this rule and the safe one does not.

Risky	`{"to": "someone@gmail.com", "subject": "Invoice", "body": "..."}`
Safe	`{"to": "ops@yourcompany.com", "subject": "Invoice", "body": "..."}`

Pick your framework

Framework	Tool name CAIN sees	Plugs into
Python (any function)	`send_email`	the function that acts
LangChain	`send_email`	your agent's tool list
LangGraph	`send_email`	the tools your graph's nodes call
OpenAI Agents SDK	`send_email`	function tools
CrewAI	`send_email`	CrewAI's before-tool-call hooks
LlamaIndex	`send_email`	FunctionTool
Pydantic AI	`send_email`	an agent capability
AutoGen (agentchat)	`send_email`	the callables you give AssistantAgent
Google ADK	`send_email`	the functions in an agent's tools
Claude Agent SDK	`mcp__ops__send_email`	PreToolUse hooks
MCP client (Python)	`send_email`	any MCP client session
OpenTelemetry	`send_email`	your existing tracing
Any other framework	`send_email`	any Python callable
TypeScript / JavaScript	`send_email`	any async function
HTTP (any language)	`send_email`	your own call site

Try CAIN-42 on your own agents

Create a free account and every new account starts with a 7-day trial of the full platform. Or try the sandbox first, with no account at all.

Create a free account → · Try the sandbox · See the whole ecosystem