 █████╗ ██╗  ██╗██████╗ 
██╔══██╗██║  ██║██╔══██╗
███████║███████║██████╔╝
██╔══██║██╔══██║██╔═══╝ 
██║  ██║██║  ██║██║     
╚═╝  ╚═╝╚═╝  ╚═╝╚═╝

Agent Handoff Protocol

The open protocol for agent-to-human ownership transfer.
Agents build. Humans claim. No signup.

Read the Spec View on GitHub

Abuse-resistantOpen specMIT License

The Problem

SaaS auth is hostile to AI agents

✗Browser-based signup flows agents can't navigate
✗Email verification, CAPTCHAs, OAuth redirects
✗Long-lived API keys that leak in agent contexts
✗Abandoned accounts when agents test and move on

The Solution

Agents build first, humans claim after

✓Agents provision sandboxes via API (no browser)
✓Build real, working output in temporary workspace
✓Human reviews actual output, not a marketing page
✓Claim transfers ownership; unclaimed sandboxes disappear

How it Works

A simple four-step flow from agent creation to human ownership

Prove

Agent solves proof-of-work challenge

Agent · ~1 sec

Build

Agent creates content in sandbox

Agent · minutes

Review

Human evaluates the real output

Human

Claim

Human takes ownership

Human · one click

→→→

Discoverable by Design

Agents find AHP-compatible services via well-known endpoints

Discovery

$ curl https://replyraptor.com/.well-known/agent-access

{
  "ahp_version": "1.0",
  "provider": {
    "name": "ReplyRaptor",
    "docs": "https://replyraptor.com/docs"
  },
  "sandbox": {
    "enabled": true,
    "admission": ["proof_of_work"],
    "ttl_hours": 48
  },
  "claim": {
    "method": "code_plus_pow"
  }
}

Create Sandbox

$ curl -X POST https://api.replyraptor.com/v1/sandboxes \
    -H "Content-Type: application/json" \
    -d '{
      "admission": {
        "type": "proof_of_work",
        "challenge": "a1b2c3...",
        "nonce": "solved_nonce"
      }
    }'

{
  "id": "sbx_7kF9xMqL2nPvR4tYwX8z",
  "public_handle": "p-7kF9xMqL2nPvR4tY",
  "agent_token": "rr_sbx_live_...",
  "expires_at": "2026-01-25T10:00:00Z"
}