Human Approval Ladder

Your agent is either too timid or too dangerous.

It asks for approval to do harmless admin work, then gets weirdly bold around actions that can cost money, damage trust, delete data, expose credentials, or post the wrong thing in public.

Human Approval Ladder gives your OpenClaw agent a practical approval system for real work.

It teaches the agent to classify actions by risk, decide whether to act now, ask first, pause for a human step, or refuse entirely, write clean approval requests instead of vague nervous questions, handle MFA, CAPTCHA, payment, consent, and account-switch moments properly, and block unsafe actions while proposing safer alternatives.

What's Included

• complete self-contained SKILL.md
• 5-tier approval ladder
• approval matrix by action class
• approval request template
• human handoff template
• refusal template
• failure pattern library and pre-execution checklist

Why this is better

• Focused on practical trust, not vague safety theatre
• Reduces both overreach and approval spam
• Works on first install with zero configuration
• Built for public posts, deletes, money, credentials, and human-gated browser flows
• Gives operators clear language they can use immediately

Works Well With

• Agent Escalation Manager
• Agent Security Hardening Kit
• Agent Operations Bundle