I Built a Local-First AI Auditor That Turns On-Chain Claims Into Receipts

Описание и применение

Автоматизированный локальный AI‑агент ClaimProof проверяет утверждения о транзакциях ERC‑8004, используя TensorFlow.js и viem, выдавая отчёт MATCH/MISMATCH с доказательствами из блоков.

Задача

Auditing on-chain claims manually is time-consuming and repetitive, requiring manual block exploration, receipt decoding, and comparison of senders and amounts.

Результат

In a real audit of FlapJax ERC‑8004 agent 364913 on BNB Smart Chain, ClaimProof returned 6 MATCH, 0 MISMATCH, 0 errors at block 128,363,575, verifying all four payout receipts and the registration receipt.

Рабочий процесс

• 1. User submits a claim to ClaimProof.
• 2. The browser trains a TensorFlow.js intent classifier (≈1 s).
• 3. The classifier maps the claim to one of eight intents (OWNER, AGENT_WALLET, URI, REGISTRATION, REGISTRATION_FILE, PAYOUT, TOKEN_FACTS, UNKNOWN).
• 4. The tool layer (viem) reads the ERC‑8004 registry, fetches tokenURI, registration files, and decodes receipts.
• 5. The verdict layer compares expected and observed values, records block and raw evidence, and generates a MATCH/MISMATCH table and Markdown report.
• 6. The user receives a report where each line can be independently re‑checked by a stranger.

Доказательства

The routing model is a two-layer TensorFlow.js classifier that trains in your browser in about a second.

Источник

The first real audit is public. FlapJax's Q31 task asked a stranger to verify its ERC-8004 identity and payouts. ClaimProof checked: the registration transaction and its block, ownerOf(364913) and getAgentWallet(364913), the stored agent URI and the registration file behind it, four FLAPJAX payouts from the treasury in the 10-12 → 10-16 window.

Источник

The result: 6 MATCH, 0 MISMATCH, 0 errors at block 128,363,575.

Источник