Create your first twin in 30 minutes

After this module you can…

• Ship one skill from idea to a verifiable receipt without help.
• Write deterministic rules that a reviewer can check without reading the prompt.

Key points

• Order that works: hatch Specialist → create Skill → assign → connect adapter → delegate.
• The manifest is the contract: SIP action, risk, approval, permissions and side effects are derived from it.
• Start in dev with a test adapter; promote to live only after one successful run.

The complete journey, step by step

The complete journey, click by click — from a brand-new account to a receipt anyone in the world can verify. Every step matches the live app; nothing here is planned or simulated.

1. 1Create your account/auth

   You do: Enter email + password, then click the confirmation link in the email.

   Behind the scenes: Auth creates your user record; first sign-in sees no Master Twin and routes you to onboarding. Google/WeChat buttons are placeholders, not live.

2. 2Hatch your Master Twin/onboarding

   You do: Three small steps: name your twin (default AiXin), pick expertise, review, press Hatch.

   Behind the scenes: A master_twins row is written (reputation 0, verified actions 0) and an ERC-8004 identity is registered on BSC Testnet — chain slowness never blocks hatching.

3. 3Create a Specialist Twin/dashboard/specialists

   You do: Click New specialist, pick a preset (Marco · Nova · Ledger · Iris) or define your own.

   Behind the scenes: Each specialist is its own specialist_twins row with its own ERC-8004 identity — individually accountable on-chain.

4. 4Install a skill (informed consent)/dashboard/skills → open a skill

   You do: Read the consent dialog — SIP action, risk level, approval need, permissions — then Accept.

   Behind the scenes: The contract is derived deterministically from the skill manifest, not the AI's mood; installing writes one skill_installs row.

5. 5Assign the skill to your SpecialistSame page — “Assigned to”

   You do: Pick your specialist from the dropdown.

   Behind the scenes: One skill_assignments row — this is scope enforcement: a twin can only use skills explicitly assigned to it.

6. 6Connect and test one adapter/dashboard/adapters

   You do: Connect what the skill needs (Gmail, Telegram, webhook, BSC) and press Test.

   Behind the scenes: Honesty guarantee: with no live adapter, execution halts as blocked / no_live_adapter with a labelled draft — never a fake success.

7. 7Delegate a task/dashboard/ask

   You do: Type your intent in plain words; answer the twin's follow-up questions; review the plan.

   Behind the scenes: Budget preflight runs first (over cap → honest refusal), your prompt is traced, a tasks row is created, then SIP validation: six deterministic rules, no AI — output is a risk level.

8. 8Decide on the Decision CardInline on Ask / Governance

   You do: Read the evidence and recommendation; Approve or Reject — a reason is recorded either way. Approving against a reject recommendation asks for a one-line override rationale.

   Behind the scenes: The decision is Ed25519-signed, its hash anchored to BSC Testnet (honest simulated/failed labels + 15-min retry queue, never a fake hash); ERC-8004 feedback written; reputation ±0.01.

9. 9Watch executionThe task view

   You do: Watch the live event log: validation → decision → execution → delivery.

   Behind the scenes: Only real tool paths run (refund ledger, CoinGecko, forecasts, Gmail, signed webhooks); every delivery attempt lands in delivery_logs; task closes done or blocked.

10. 10Verify — and let anyone verify/verify/<sip_id>

    You do: Open the receipt link and send it to a friend — it works without logging in.

    Behind the scenes: Public page: SIP ID, payload hash, Ed25519 signature + public key, BscScan link, block number, ERC-8004 txs — personal data redacted. The page no unverified agent platform can show.

Getting it wrong is fine — the explanation tells you why.