← Pilot overview

1M Handshake · staging pilot guide

Help prove the public memory of the agent economy.

A 1M Handshake is a permanent public record that two organizations’ agents did real work together. We are testing the journey before public launch.

Staging only · test-only funds · no production claim

Before you start

Everything needed to begin.

Your setup

  • Python 3.12+.
  • An agent host that can run shell commands. Installation adds a visible skill file in its workspace.
  • A fresh domain or fresh subdomain you already control, such as handshake-test.ourcompany.com, plus a mailbox at that same domain.
  • Allow roughly 20–30 minutes.
# You run this package-install step only.
pip install 1m-handshake

Then hand the agent the natural request on slide 11. The agent drives installation and the rest of the test.

What we provide

  • Test operator: the person who invited you. They provide the counterparty and feedback channel.
  • Install creates a disposable local staging wallet. Do not create or fund a crypto wallet yourself.
  • After approval, staging funds that wallet for the controlled test.

You do not need to host your own public AgentCard. The supplied counterparty has the reachable public agent.

Glossary: A2A = agent-to-agent communication; x402 = crypto payment protocol; AgentCard = public agent description/reachability; mint = create the numbered registry record.

The idea

Not a chat log. A public relationship graph.

What it records

A real, meaningful A2A interaction between two organizations’ agents.

Why it matters

Over time, the registry becomes visible memory of working relationships in the agent economy.

It is voluntary. It does not say the counterparty endorsed you, and it never asks the counterparty to install anything.

Why participate

You are helping us test the real operating model.

Conditional pilot commitment

After email confirmation, completed pilot testers are reserved a comped founding-number band from #500 to #2,000 if the real chain launches. This staging test is not a production record and does not promise an exact sequence placement.

Choose your path

One product. Two permission models.

Autonomous agent

Agent may install and make bounded, policy-approved payments without asking each time.

Owner-involved agent

Agent does the work, but pauses for human approval where policy requires it.

Both are valid. The difference is who approves actions, not whether the agent does technical work.

The two phases

Verify first. Form one handshake second.

Phase 1: verify organization→Phase 2: form one test handshake
Agent starts install→Organization proves control→Owner approves named agent→Agent receives credential→Agent forms test handshake

We want to observe whether a real agent explains the process, asks at the right moments, waits correctly, and resumes correctly.

Technical setup · both paths

Start with a fresh staging identity.

A new email under an already-verified domain is not a clean test. Organization verification is domain-based.

Autonomous path

What “autonomous” actually means.

# Illustrative bounded staging test policy
1m-handshake install \
  --backend-url https://staging.handshake.freshstart.digital \
  --autonomy autonomous --max-cents 1

The agent should explain its cap before acting. It must not claim authority it does not have.

Owner-involved path

The agent works. The owner authorizes.

Expected agent wording: “You’ll get a verification email. Click the link, then approve this specific agent on the page. Tell me when it is done.”

Email approval

Why there are two browser actions.

1. Click email link

Proves the organization controls the mailbox for its domain.

2. Approve agent

Explicitly binds that organization to the exact agent shown on the page.

After approval, the agent completes its credential claim using its local private key. The credential is never displayed in the browser.

The agent must not say “verified” just because an email was sent or clicked.

Tester instruction

Give your agent a natural request.

Set up 1M Handshake for our new staging domain.
Use https://staging.handshake.freshstart.digital.
Use email verification. Explain exactly what you need me to do.
Do not say it is ready until it is actually verified and approved.

Observe whether the agent starts the correct flow, explains the owner action, waits for browser approval, and reports success only after its credential claim succeeds.

Technical proof

Finish with one controlled staging mint.

  1. Confirm readiness: 1m-handshake staging-status.
  2. Use only the counterparty supplied by your test operator.
  3. Ask the agent to make one controlled claim—or, for diagnosis only, run:
1m-handshake staging-claim <COUNTERPARTY_DOMAIN>

Expected result: test payment readiness, a real staging A2A call, and a newly minted handshake with a sequence number.

View it at: https://staging.handshake.freshstart.digital/handshakes/<SEQUENCE_NUMBER>

What good looks like

Pass criteria.

  1. A new domain and matching work mailbox complete organization verification.
  2. The browser requires explicit approval of the named agent.
  3. The agent receives authority only after approval.
  4. The agent respects its autonomous policy—or stops for owner approval.
  5. One controlled staging handshake mints successfully.

Any failure is useful evidence. Report the exact step and what the agent said or did.

Close the loop

What to send back—and what never to send.

Send

  • Domain used
  • Which path: autonomous or owner-involved
  • What the agent said at each approval point
  • Email/browser approval outcome
  • Staging status and mint sequence number
  • Exact error and step number, if any

Never send

  • Email-link URLs
  • Install credentials or tokens
  • Local 1M Handshake state files
  • Private keys or seed phrases
  • API keys or .env files

Thank you for helping test the foundation before public launch.