elizaos-plugin-yieldsignal

Real-time, risk-weighted yield signals — paid per call ($0.01 USDC) via the x402 protocol through YieldSignal:

• ETH liquid staking (Ethereum mainnet) across Lido, Rocket Pool, Coinbase Wrapped Staked ETH, Frax Ether and Binance Staked ETH
• USDC / WETH lending (Base) across Aave, Compound, Morpho, Moonwell, Euler and Fluid

Per-asset verified accuracy is public and free at /accuracy.json, computed from an on-chain EAS track record — check it before trusting the signal.

Verifiable, not self-declared. Responses are EIP-712 signed and the seller keeps a public on-chain track record (ERC-8004 identity + EAS attestations on Base). Its machine-readable accuracy is live at /accuracy.json — currently 93.75% within-tolerance (the flagged protocol was the leader or within 25bps), average regret 8bps. A paying agent can check who to trust by the proven record, not the promise.

Five products, not one

The signal answers what pays best right now — which is a commodity anyone gives away. The four risk reports answer questions nothing else sells, and each one names what it could not establish instead of filling the gap with a guess.

| Action | Question it answers | |---|---| | GET_YIELD_SIGNAL | What pays best right now (Base lending or ETH staking) | | GET_YIELD_DURABILITY | Is this yield real, or a promotion about to end? | | GET_EXIT_CAPACITY | Can I actually withdraw my size from that market? | | GET_RATE_SENSITIVITY | How close is this market to the kink where borrow rates explode? | | GET_SHARED_EXPOSURE | I'm in N venues — but behind how many distinct risks? |

Measured on live readings:

• Durability — WETH on Base led with euler at 299bps, but 57.9% of that was incentive (floor 126bps), while aave's 153bps was entirely base interest. Without incentives the ranking flips.
• Sensitivity — Compound's USDC market sat 0.17 percentage points from its kink, where borrowing goes from 4% to 16%. Hours later it crossed, and the cost of borrowing rose 1.7x.
• Exposure — a portfolio across three venues had 81% of its traceable capital behind one collateral, reaching it through two of them. The apparent diversification was not real.

The four reports are Base lending only (USDC / WETH): liquid staking has no utilisation, no interest-rate curve and no itemised incentive to decompose, so the plugin refuses those assets rather than letting the agent pay for a 404.

Quick start: see examples/ for a copy-paste plugin setup and a standalone fetch-and-verify snippet.

Purpose / role

Adds a single buyer-side action, GET_YIELD_SIGNAL, that lets an elizaOS agent proactively call and pay for an external x402-protected API. This is distinct from @elizaos/plugin-x402 (seller-side middleware for protecting this agent's own HTTP routes) — there's currently no overlap, since that package registers no actions/providers/services of its own.

Every response from YieldSignal is signed (EIP-712 typed data) by the payment-receiving address, which also holds an on-chain ERC-8004 agent identity (/agent-card.json) and periodically publishes EAS attestations of past readings on Base mainnet (/track-record).

Plugin surface

| Kind | Name | What it does | |------|------|--------------| | Action | GET_YIELD_SIGNAL | Fetches the current best protocol and risk-weighted APY (bps). Parses ETH_STAKING/USDC/WETH from the triggering message text — staking wins on "stake"/"staking"/bare "ETH", WETH and USDC on an explicit mention, defaulting to USDC. |

Layout

src/
  index.ts                     Plugin entry; exports yieldSignalPlugin
  client.ts                    x402 payment + fetch (CdpX402Client + @x402/fetch)
  actions/
    get-yield-signal.ts        GET_YIELD_SIGNAL action handler
  get-yield-signal.test.ts     Unit tests (no network — validates registration/shape)

Config / env vars

| Env var | Required | Purpose | |---------|----------|---------| | CDP_API_KEY_ID | Required | Coinbase CDP API key ID — provisions the plugin's own payment wallet | | CDP_API_KEY_SECRET | Required | Coinbase CDP API key secret | | CDP_WALLET_SECRET | Required | Coinbase CDP wallet secret |

The plugin's wallet needs a small amount of USDC on Base ($0.01 per call).

Usage

import { yieldSignalPlugin } from "elizaos-plugin-yieldsignal";

const character = {
  // ...
  plugins: [yieldSignalPlugin],
};

Security posture

This plugin holds a spending wallet, so it enforces — client-side — the guarantees the service advertises, rather than trusting the server:

• Hard spend policy (via the CDP SDK's spend controls, buildSpendControls): a per-call cap ($0.10 USDC, well above the advertised $0.01/$0.05), a rolling 24h cumulative cap ($2.00), and fixed allowlists for network (Base / eip155:8453), asset (USDC) and payee (0x5611…472a). A tampered or swapped 402 challenge cannot make the wallet pay more, a different asset/chain, or a different recipient.
• EIP-712 verification (verifyYieldSignalSignature): every response's X-Signal-* headers are verified with viem.verifyTypedData, the embedded contentHash is checked against keccak256(rawBody), and the signer is pinned to the advertised payee. An unsigned or tampered response is rejected, not trusted.
• Runtime schema validation (parseYieldSignalResponse): the body is validated against the expected shape, never blindly cast.
• Bounded I/O: the paid fetch aborts after a timeout (default 15s), so a hung server can't block the agent.
• Explicit-intent gate: validate only returns true when the message actually expresses a yield/rate intent (hasYieldIntent) — the wallet is never touched on unrelated messages.

Adversarial unit tests cover signature rejection (wrong signer, tampered body, bad signature, malformed payload), schema rejection, the spend-policy values, and the intent gate. See src/security.test.ts.

Why a dedicated wallet instead of the agent's own signer?

x402 payment signing (@x402/fetch + @coinbase/cdp-sdk/x402) is independent of whatever wallet/signing setup the agent otherwise uses — this plugin pays for the call the same way it's already proven against production (see scripts/testPaidCall.mts in the service's own repo), rather than adapting a second signing path into the agent's primary wallet integration.