SShortSingh.
Back to feed

Set up an AI agent to request human approval before risky actions

0
·2 views

Coding agents can deploy code, send emails and delete data without supervision, which creates risk. This tutorial shows how to configure an agent to pause before such steps, ask a designated person for approval, and proceed only once that person agrees. The setup uses configuration and plain-language rules rather than custom code, and takes about ten minutes. The approver receives an email, reviews the request, and approves or rejects it with a one-time code or passkey, after which the agent continues or stops. It requires an MCP-compatible agent such as Claude Code, Cursor, VS Code with GitHub Copilot, or Codex, plus a free Guizzo account.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Guide to self-hosting n8n on a VPS with Docker and Caddy

The article explains how to run n8n, a workflow automation tool, on a single Linux VPS for data control and no per-execution fees. It walks through the setup using two Docker containers, one configuration file, and automatic HTTPS. Prerequisites include Ubuntu 24.04, root SSH access, a controlled domain, and sufficient server memory, with 8 GB suggested for personal or small-team use. The design keeps n8n reachable only locally while Caddy handles public traffic and free certificates. The guide also notes that the Community Edition's licence restricts certain uses, such as hosting for clients.

0
ProgrammingDEV Community ·

Lawra Launches Anonymous Social Platform Focused on Free Expression

Lawra is an independent social networking project that offers an anonymous alternative to traditional platforms built around personal profiles. It combines public posting, threaded comments, public message rooms, private messaging, and media sharing in one service. Instead of centering users' names and photos, the platform emphasizes the ideas people want to express and the conversations they join. Posts can run up to 1,500 words and can be published without a public identity. The project's stated premise is that a user's voice should matter more than who they are.

0
ProgrammingDEV Community ·

Developer's Test Finds Zero-Knowledge Circuits in Agent Games Reject Falsified Proofs

A developer tested the zkVerify Agent Games arena circuits by attempting to submit dishonest proofs, working from official circuit artifacts downloaded on 21 November 2026. Using snarkjs locally, they generated Groth16 proofs for a payment-approval Gatekeeper circuit and found that claiming APPROVE when the balance fell below the minimum produced an assertion failure. No proof was generated, and the refusal occurred within 301 milliseconds. The write-up stresses that it is a local verification report, not a live tournament run, and that no funds or mainnet submissions were involved. The circuit also certifies correct rejections, not only approvals, according to the author.

0
ProgrammingDEV Community ·

Two x402 Payment Rails Diverge: AWS Invoice Billing Versus Wallet-Signed USDC

On November 19, AWS Marketplace announced agentic payments over x402 through Bedrock AgentCore, letting agents pay per request in US dollars on the customer's existing AWS bill. The protocol's core flow, which returns an HTTP 402 challenge that the client answers with a payment, can run on either rail. The two systems differ mainly in who may pay: wallet holders using USDC on Base versus AWS account holders buying from verified Marketplace sellers. The author's platform uses wallet-signed USDC, which lets agents transact without accounts or seller onboarding. The article argues that an agent must check what kind of payer identity and settlement a 402 requires, since 'supports x402' does not guarantee it can actually pay.