Agent Guides

The Problem: Financial Agents Without Audit Trails

On February 20, 2026, Uniswap Labs released 7 Agent Skills — structured actions that give AI agents direct access to core Uniswap protocol functions: swaps, LP positions, routing, and more.

This is a major milestone for onchain agentic finance. For the first time, AI agents can execute real DeFi operations programmatically, without human hand-holding.

But there's a gap: financial agents acting onchain with no neutral audit trail.

When an agent makes a trade, the blockchain records the transaction. But it doesn't record:

For institutional use, regulatory compliance, or DAO treasury management, this gap matters.

The Stack: Uniswap Skills + CustosNetwork

Combine Uniswap Agent Skills with CustosNetwork inscriptions to get a complete agentic finance + accountability stack:

The pattern: Inscribe *before* executing. Hash your decision rationale, write it onchain, then call the Uniswap Skill. The inscription timestamp proves the reasoning preceded the trade.

Integration Pattern

### Step 1 — Inscribe the decision (before the trade)

import { createWalletClient, createPublicClient, http, keccak256, toBytes } from "viem";
import { base } from "viem/chains";

const PROXY = "0x9B5FD0B02355E954F159F33D7886e4198ee777b9";
const USDC  = "0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913";

// Build your decision payload before trading
const decision = {
  action: "swap",
  tokenIn: "USDC",
  tokenOut: "ETH",
  amountIn: "1000",
  rationale: "ETH/USDC at 30d low; rebalancing per allocation policy",
  timestamp: Date.now(),
};

const content   = JSON.stringify(decision);
const proofHash = keccak256(toBytes(content));

// Approve + inscribe BEFORE calling the Uniswap Skill
await walletClient.writeContract({
  address: USDC, abi: erc20Abi, functionName: "approve",
  args: [PROXY, 100_000n], // 0.1 USDC inscription fee
});

await walletClient.writeContract({
  address: PROXY, abi: proxyAbi, functionName: "inscribe",
  args: [proofHash, prevHash, "market", "swap USDC→ETH: rebalancing per allocation policy"],
});

// THEN execute via Uniswap Agent Skill
const swapResult = await uniswapSkill.swap({ ... });

### Step 2 — Inscribe the outcome (after execution)

const outcome = {
  action: "swap_complete",
  txHash: swapResult.txHash,
  amountIn: swapResult.amountIn,
  amountOut: swapResult.amountOut,
  slippage: swapResult.slippage,
  decisionHash: proofHash, // links back to pre-trade inscription
};

await walletClient.writeContract({
  address: PROXY, abi: proxyAbi, functionName: "inscribe",
  args: [keccak256(toBytes(JSON.stringify(outcome))), proofHash, "market", "swap complete: got X ETH for Y USDC"],
});

Why "Inscribe Before" Matters

The inscription timestamp on Base is immutable. If you inscribe the rationale *before* the swap executes, you've proven the decision was formed before the trade. This matters for:

If you inscribe *after*, the timestamp can't prove the reasoning preceded the action.

The Audit Gap: Uniswap Skills Alone

Uniswap Agent Skills record execution onchain — the swap transaction is permanently on Base. But execution records only show *what happened*, not *why*.

An auditor looking at a sequence of agent trades can verify amounts and prices. They cannot verify:

CustosNetwork fills this gap. The prevHash chain-link means auditors can walk the entire decision history, verify continuity, and detect missing cycles.

Uniswap Agent Skills Available (Feb 2026)

Each Skill maps to a structured onchain action. Each action is a candidate for a CustosNetwork inscription.

Production Notes

*Based on Uniswap Agent Skills release Feb 20, 2026 and CustosNetwork V5 proxy. Both live on Base mainnet.*