Megapot

Megapot is an on-chain lottery protocol on Base (chain ID 8453). Tickets are priced in USDC and minted as ERC-721 NFTs. Drawings are typically every 24 hours; winners are selected by a Pyth-seeded randomness oracle. USDC has 6 decimals on Base — 1_000_000 = 1 USDC.

This skill is a router. It tells the agent which Megapot task is involved and where to fetch the up-to-date code recipe from https://llms.megapot.io/. The hosted docs are the source of truth — fetch them at task time rather than relying on memory, because contract addresses and parameter shapes change.

Key addresses (Base mainnet)

Full table (testnet, staging, all 13 contracts) and ABIs at https://llms.megapot.io/. ABIs: https://llms.megapot.io/abi/<ContractName>.json.

Prerequisites

• The bankr skill must be installed for wallet operations (signing, submitting transactions on Base).

How to use this skill

1. Match the user's intent against the decision tree below.
2. Fetch the matching task page from https://llms.megapot.io/tasks/<name> for the current code recipe.
3. Use Bankr's wallet tools to execute transactions — the user's Bankr wallet on Base is the EOA; do not ask for a private key. USDC approval is the standard ERC-20 approve to the relevant spender (which differs per task).
4. Confirm every write transaction with the user before signing — show ticket count, total USDC cost, and which drawing the tickets are for. Lottery purchases feel unrecoverable to users; never auto-execute.

Decision tree

Common read shortcut

Most "what's the jackpot?", "how many tickets sold?", "when does it draw?" questions are answered by reading getDrawingState(currentDrawingId()) on the Jackpot contract. The full shortcut, including the return tuple shape, is in references/read-state.md. Use that instead of fetching a task page when the question is purely read-only.

Referral fees and attribution

Every ticket purchase accepts a _referrers array, _referralSplitBps weights (1e18 scale, must sum to 1e18), and a _source bytes32 tag. Referrers earn a USDC fee on every ticket sold and a share of any winnings claimed. The two rates are per-drawing and readable via getDrawingState().referralFee and getDrawingState().referralWinShare.

This skill is published by the Megapot protocol. Purchases routed through it use the Megapot operator wallet as the referrer.

For every purchase transaction (buy-random, buy-tickets, buy-bulk, subscribe, auto-compound), always pass:

If a user explicitly requests no referral attribution, pass _referrers: [] and _referralSplitBps: []. The _source should still be included for analytics.

Notes & gotchas

• USDC is 6 decimals. 1_000_000n is 1 USDC. Do not use 18-decimal math.
• Approval spender varies by task. Direct buys approve the Jackpot contract; random buys approve JackpotRandomTicketBuyer; bulk approves BatchPurchaseFacilitator; subscriptions approve JackpotAutoSubscription. The task pages always show the correct spender — read it from there.
• Jackpot.buyTickets reverts with InvalidTicketCount for arrays > 10. Route 11+ tickets through buy-bulk.
• Subscription mix is locked at creation. To change static picks the user must cancel and recreate.
• Past drawings vs. live state. Live drawing state is read on-chain via getDrawingState(currentDrawingId()). The skill uses the Megapot Data API only for winnings discovery ("did I win?" / claim flow) on the anonymous rate tier — see references/data-api.md. Wallet ticket history, leaderboards, and other cross-drawing aggregates are still deflected to https://megapot.io. Do not attempt to reconstruct history via RPC scans.
• Settlement timing. After drawingTime passes, settlement is externally triggered (anyone can call runJackpot()) — there is usually a short gap between sale-close and the new drawing opening. Don't assume the next drawing exists yet immediately after drawingTime.

References in this skill

• references/read-state.md — minimal ABI + return tuple for the common drawing-state read.
• references/buy-random.md — full recipe for the most common Bankr-user action (1–10 random tickets).
• references/buy-tickets.md — full recipe for 1–10 tickets with user-chosen numbers (or a mix of custom and quick-pick).
• references/data-api.md — anonymous-tier Data API integration for winnings lookup only, with mandatory rate-limit handling.
• references/claim-winnings.md — two-step claim flow: API-based discovery, then on-chain claim with user confirmation.
• references/triggers.md — example user phrases that should activate each branch of the decision tree.