Human-in-the-loop patterns for AI agents

This page is the canonical guide to human-in-the-loop approval patterns for TypeScript AI agents. For a copy-paste recipe and UI wiring, see HITL approvals and ToolConfirmation.

#Gate a tool

import { makeTool } from '@agentskit/tools'

const deployTool = makeTool({
  name: 'deploy',
  description: 'Deploy to production',
  schema: z.object({ service: z.string() }),
  requiresConfirmation: true,
  execute: async ({ service }) => deploy(service),
})

Runtime pauses on invocation and emits tool.awaiting-approval. Resume with chat.approve(toolCallId) or chat.deny(toolCallId, reason).

#UI

• ToolConfirmation — drop-in React / Vue / etc.

#Patterns

• Auto-approve low risk: approve if cost under threshold; gate the rest.
• Review queue: persist awaiting-approval to a DB; humans approve from dashboard.
• Double-sign: require two approvers; track via shared context.

#Related

• Recipe: HITL approvals
• Recipe: confirmation-gated tool
• Security → mandatory sandbox