Approve or block 1 agent spend in 60 seconds with scoped smart-account permissions, 1Shot x402 relay packet, Venice-compatible risk note, and an audit receipt.
What it does
Smart Spend Sentinel lets a wallet owner approve or block one AI-agent spend request in about 60 seconds. The workbench takes seller, resource, category, cadence, amount, risk, and expiry inputs. It returns either an APPROVED scoped permission package or a BLOCKED receipt with clear reasons.
How it uses the sponsor stack
MetaMask Smart Accounts Kit is represented as the permission model: scoped caveats, expiry, max amount, cadence, and a visible typed-data authorization step. 1Shot is represented in the package inspector as the Permissionless relay path and x402 payment request preview. Venice AI is used as the risk-enrichment path when VENICE_API_KEY is configured, with an honest local risk note fallback when it is not.
Proof surfaces
The live app has three judge-facing surfaces: /app for approve/block interaction, /app/package for the relay packet and permission package, and /app/audit for saved receipts. The demo video shows wallet connect, signature proof, approval, blocked compute-spike case, and the package inspector.
Built during the hackathon: deployed Next.js app on Cloudflare Workers, deterministic policy evaluator, smart-account permission package builder, 1Shot x402 relay preview, Venice-compatible risk note, package/audit pages, mobile and desktop Playwright tests.
Current limitation: live 1Shot relayer submission requires production credentials, so the demo shows the honest package/receipt that would be sent instead of faking a transaction hash.
No fundraising. This is a hackathon prototype focused on wallet-owner control for AI-agent payments.