Recipes

Guard refunds

For a support agent that issues refunds, in 15 frameworks.

Last reviewed 2026-10-02

Routine refunds stay automatic; large ones are held for a human, with the order and amount in the approval.

The rule

{
  "name": "refunds-guard",
  "effect": "require_approval",
  "priority": 10,
  "match_path": "/tools/issue_refund",
  "conditions": [
    {
      "field": "amount_usd",
      "op": "gt",
      "value": 50
    }
  ]
}

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

Risky	`{"order_id": "ord_1042", "amount_usd": 480}`
Safe	`{"order_id": "ord_1042", "amount_usd": 12}`

Pick your framework

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