What's inside

One SKILL.md (the playbook) + one starter template. A preview:

Proof-of-Humanity Checkout

When to use

• You're letting an AI shopping or booking agent spend real money for users.
• Chargebacks or fraud reviews keep asking "was this the customer or their bot?"
• You need the checkout to prove a verified human authorized this specific purchase — not just that an account was logged in.
• You're evaluating identity-verification vendors for agent-driven commerce.

Workflow

1. Define the delegation (input: what the agent may buy → output: delegation contract)
   Write the contract between human and agent: spending cap per order and per day, allowed merchants/categories, what needs explicit approval (anything over $X, subscriptions, non-refundable items), and expiry (delegation lasts a session, a day, a trip — never forever). The agent presents this contract, not just an identity token, at checkout.
2. Choose the proof mechanism (input: contract → output: proof design)
   Pick how "a verified human authorized this" is proven: delegated identity proofs the agent carries (proof-of-human, proof-of-presence), step-up authentication for over-cap orders (biometric or device approval on the human's phone), or cryptographic receipts (the human's device signs the order hash). Match the mechanism to the stakes: groceries get delegation, wire transfers get step-up + signature. Never rely on "the agent said the human approved" — that's the claim, not the proof.
3. Design the approval moment (input: proof design → output: approval UX)
   For over-cap or out-of-contract orders, the human sees: what, from whom, for how much, arriving when, and what happens on cancel — on their own device, in under 10 seconds, with approve/deny/edit. Default to deny on timeout. Log the approval with the proof attached; the receipt must show it later.
4. Build the receipt (input: order + proof → output: receipt spec)
   Every agent-placed order produces a receipt showing: the human principal, the agent that acted, the delegation contract version, the proof presented, the approval record (or "within standing delegation"), and the merchant's view (what the merchant saw — they should see a verified-human-backed order, not raw personal data). Receipts are the evidence when the human says "I didn't order that."
5. Handle the disputes (input: receipt spec → output: dispute playbook)
   "I didn't authorize this" → pull the receipt: was there a valid delegation, a proof, an approval? If yes, show the human their own approval record (most disputes end here). If the proof is missing or the delegation expired, refund and treat it as a system failure, not user fraud. Track dispute rate per agent version — a rising rate is a delegation-design bug.
6. Mind the privacy line (input: full design → output: data-minimization review)
   The merchant needs to know the order is human-backed, not who the human is: proofs should reveal authorization, not identity attributes. Review what personal data crosses to the merchant, the agent provider, and the identity verifier — and cut everything the checkout doesn't strictly need.

Template preview

Proof-of-Humanity Checkout Worksheet

The question every agent checkout must answer: how does the other side know a real human authorized this?

1. Delegation contract (human → agent)

• Per-order cap:
• Per-day cap:
• Allowed merchants / categories:
• Always needs explicit approval (over $X, subscriptions, non-refundable):
• Delegation expiry (session / day / trip — never forever):

2. Proof mechanism (match to stakes)

Never "the agent said the human approved" — that's the claim, not the proof.

3. Approval moment (over-cap / out-of-contract)

What the human sees on their own device (what / from whom / how much / arriving when / cancel terms):

Buy the full skill pack — $4.99 Pro — all 511 skills