Yield tools for Zerion CLI

Seven vaults.fyi skills are now live in Zerion CLI for coding agents. The vaults.fyi skills package yield research, deposit flows, rebalance flows, and risk monitoring functions into commands an agent can run alongside Zerion's wallet primitives.

We are excited to work with vaults.fyi as our early yield partner because of their data depth and transaction-building capabilities. The collaboration between our products unlocks workflows neither side could deliver alone.

— Evgeny Yurtaev, Co-founder & CEO at Zerion

About Zerion AI

Zerion CLI exposes wallet primitives an agent can use directly: portfolio reads, signing, swaps, bridges, and policy-scoped trading credentials across EVM chains and Solana. Skills install once and load on demand across 20+ supported agents (Claude Code, Cursor, Codex, Gemini CLI, and others).

Zerion is offering free developer plans for a limited time for builders on the CLI. Go to zerion.io/agents to get started.

Why vaults.fyi as an initial yield partner

To be useful as an initial yield partner, three things had to be in place on the vaults.fyi side. They already were, via the MCP:

• Coverage. 1,000+ vaults across 20+ networks, refreshed continuously. The same skills work whether the user is in Aave, Morpho, Spark, Euler, Fluid, or any curator product that sits on top.
• Agent-native data shape. Multi-window APY (1d, 7d, 30d) with base and reward decomposition. Composite APY for yield-bearing deposit assets like sUSDe and sUSDS, so nested yields are not undercounted. Per-network benchmark rates so an agent can answer "is this above market" in one call. Reputation scores and curator attribution on every vault.
• Decoded transaction building. The vaults.fyi MCP returns viem-compatible unsigned transactions with a stepDecodings array (what each step is doing) and explicit signing instructions.

Those three pieces are what made the skill scope this broad. Each one removes work the skill author would otherwise have to do themselves.

The seven skills

All seven are MIT-licensed and live at github.com/zeriontech/zerion-ai under skills/zerion-vaultsfyi-*.

Walkthrough: yield optimizer

Here is the zerion-vaultsfyi-yield-optimizer skill end to end. It touches every layer: Zerion's portfolio view, vaults.fyi's yield data, benchmark context, and the action layer.

Prompt:

"Audit my wallet 0xabc… for yield opportunities. Where am I leaving money on the table?"

What the agent runs:

1. Portfolio snapshot from Zerion. zerion analyze 0xabc… returns the broad picture: every token balance, every DeFi position, USD values, and PnL. This covers protocols beyond vaults too (DEX LPs, staking, lending), so the audit is not blind to non-vault context.

2. Yield layer from vaults.fyi. Three MCP calls in parallel:

• positions returns the wallet's vault positions with APY, score, flags, and curator data
• wallet_balances returns idle tokens not deployed in any vault
• rewards returns claimable vault rewards

3. Benchmark context. For each network in the wallet, benchmark_apy with code: "usd" and code: "eth" establishes whether current yields are above or below market. (Benchmark = TVL-weighted average APY across the five largest vaults in that denomination on that network, $1M minimum each. Background: our benchmarks post.)

4. Vault search for trailing positions and idle balances. For each position trailing its benchmark, vaults_search finds alternatives sorted by 30-day APY (excluding active critical flags, minimum $5M TVL, same deposit asset). Composite APY is used when the deposit asset is itself yield-bearing (sUSDe, sUSDS). For significant idle balances, best_deposit_options ranks vaults by risk-adjusted yield.

5. Compiled audit. Output ordered by what the user can act on first: risk alerts, claimable rewards, opportunity cost, idle-asset suggestions, concentration warnings, portfolio summary.

6. Action confirmation. A prioritized action list. The skill explicitly asks before executing anything. If the user approves a rotation, zerion swap (or zerion bridge --cheapest cross-chain) handles asset acquisition, and the vaults.fyi MCP builds the deposit transaction. Decoded steps surface before signing.

Sample output (illustrative; numbers vary by wallet):

Risk alerts:
  none

Claimable rewards:
  142 MORPHO on Base (~$220)

Opportunity cost (annualized):
  $18,400 in Aave v3 USDC mainnet (3.2% 30d)
    → Steakhouse USDC Morpho mainnet (5.1% 30d, Score 81)
    Gap: 1.9 pts, ~$350/yr

Idle:
  4,200 USDC on Base
    → Smokehouse USDC (6.0% 30d, Score 81)

Concentration:
  68% of deployed value in a single Morpho vault

Suggested actions:
  1. Claim 142 MORPHO
  2. Rotate $18,400 from Aave v3 USDC → Steakhouse USDC
  3. Deploy 4,200 USDC on Base into Smokehouse USDC
  4. Reduce single-vault concentration after rotation

Numbers vary by wallet. The structure is consistent: one prompt produces an audit covering risk alerts, claimable rewards, opportunity cost in USD, idle-asset suggestions, and a confirm-then-execute action list. The skill does not pick for the user. It surfaces options and waits.

Safety: two independent boundaries

When an agent can move funds on a user's behalf, the safety bounds matter as much as the data does. Two boundaries combine here, and neither depends on the other.

Zerion's manual-vs-agent split. Wallet creation, key import, agent-token issuance, and policy attachment are all manual and passphrase-prompted. Once an agent token is minted and scoped (chain allowlist, contract allowlist, expiry, transfer caps), the agent operates inside that policy. No spending credential widens without the user in the loop. The token works like an API key with spending power, scoped down to what the policy allows.

vaults.fyi flag gating. Every vault payload carries a flags[] array with severity. The vaults.fyi skills are written to act on it: the deposit and rebalance skills refuse vaults with active critical flags and require explicit user approval on warning flags. Flag propagation across curated vault systems is automatic, so when an upstream incident hits (a Kelp rsETH exploit, a Resolv bad-debt event), every downstream vault holding the affected asset lights up too, with reputation-score penalties already applied.

The two boundaries work together. The agent token sets what is allowed (chain, contract, expiry, transfer caps), enforced by the policy. The skills set what is prudent (no critical-flagged vaults, explicit approval on warnings), enforced by how the skills are written. The loop can run unattended without expanding what the agent can actually do to the wallet.

Getting started

Three steps.

1. Install the Zerion CLI.

npx -y zerion-cli init -y --browser

This installs the CLI globally, opens the Zerion dashboard for an API key, and installs the full skill set across detected coding agents.

2. Connect the vaults.fyi MCP. Same hosted endpoint and config as in the MCP launch post. If you already have it set up, skip this step.

3. Try a prompt. A few that exercise the new skills:

• "Audit my wallet 0x… for yield opportunities." (yield-optimizer)
• "How does the USDC yield market look on Base right now?" (market-intel)
• "Move my USDC from this Morpho vault to a better one on any chain." (rebalance)
• "Watch this wallet and tell me about its yield strategy." (watchlist)

The agent loads the right skill, runs the underlying Zerion CLI commands plus the vaults.fyi MCP tools in sequence, and surfaces decoded steps before signing.

What's next

A few directions we're already working on with Zerion:

• More skill compositions. Specifically: yield-bearing-asset routing (sUSDe, sUSDS) where the deposit asset is itself a yield primitive, and multi-vault portfolio rebalances that respect concentration policies.
• x402 pay-per-call. Both sides expose x402 endpoints. We are working through what a bundled cost model looks like for autonomous loops that do not want to manage API keys.

Questions, feedback, or interest in a custom skill: @vaultsfyi on X, or DM us.