feat: add yield — multi-chain yield optimization via yield.xyz

[tonyagents]

Skill name

yield-optimization

Description

Multi-chain yield optimization via StakeKit (yield.xyz) — find the best APY opportunities across lending protocols, liquidity pools, and staking on 75+ chains. Agent finds rates, builds unsigned transactions, and your MoonPay wallet signs and executes them.

Primary chain

Multi-chain: Ethereum, Polygon, Arbitrum, Base, Solana, and more

Primary token

USDC, ETH (varies by yield opportunity)

Checklist

• Frontmatter (name, description, tags) present
• Skill registered in .claude-plugin/marketplace.json under yield-skills plugin at skills/yield-optimization/
• Naming convention: yield-optimization/ (partner=yield, name=optimization)
• Shell scripts verified in stakekit/yield-agent GitHub repo (scripts/ directory)
• clawhub confirmed real (npm v0.8.0)
• mp transaction send confirmed real
• No Python code — pure shell scripts

MoonPay Integration

StakeKit builds unsigned transactions; the agent signs and broadcasts them via mp transaction send --wallet yield-agent --chain --transaction .

Example Usage

# Install
npx clawhub@latest install yield-agent

# Find best USDC yields on Base
./scripts/find-yields.sh base USDC

# Enter a position (returns unsigned tx)
./scripts/enter-position.sh base-usdc-aave-v3-lending 0xYOUR_ADDR '{"amount":"100"}'

# Sign and execute with MoonPay
mp transaction send --wallet yield-agent --chain base --transaction '<unsigned-tx>'

# Monitor position
./scripts/check-portfolio.sh base-usdc-aave-v3-lending 0xYOUR_ADDR

[kevarifin14]

PR Review: skills/yield/ — Multi-Chain Yield Optimization

Naming Convention Violation (Blocking)

The skill is located at skills/yield/ but CLAUDE.md is explicit: partner skills must follow skills/{partner}-{name}/ naming (e.g., corbits-marketplace, dune-analytics). Since this is a yield.xyz integration, it should be skills/yield-optimization/ or, more consistent with the convention, skills/yieldxyz-optimization/ or similar. The current bare yield name has no partner prefix and does not follow either the MoonPay (moonpay-{name}) or partner ({partner}-{name}) pattern.

The name field in the YAML frontmatter (name: yield) should match the directory name once corrected.

marketplace.json Placement (Blocking)

The new skill ./skills/yield is appended inside the moonpay-skills plugin block. This is a partner skill, not a MoonPay skill. Either:

1. It should be in its own plugin entry (like the existing pattern suggests partner plugins could be separate), or
2. At minimum, this needs discussion — lumping a third-party partner's skill into the moonpay-skills plugin misrepresents ownership.

Hallucinated / Unverifiable CLI Commands (Blocking)

Several mp commands used in this skill do not appear in any existing skill in this repo and appear to be fabricated:

• mp transaction send --wallet ... --chain ... --transaction ... — No existing skill uses mp transaction send. The only transaction-related command in the repo is mp transaction list. This is the entire foundation of the skill's workflow and it appears to be hallucinated.
• mp wallet add-ledger --name "yield-ledger" — The moonpay-hardware-wallet skill should be checked to confirm this command, but the flag pattern looks invented.
• mp deposit create — Not found in other skills.
• mp virtual-account offramp create / mp virtual-account onramp create — Need verification against the actual moonpay-virtual-account skill.

If mp transaction send is not a real command, the entire skill is non-functional. This must be verified before merge.

Python Code in an Instructional Skill (Major)

Skills are explicitly "not code — they are instructional guides. All execution happens through CLIs." This skill contains substantial Python code blocks:

• from yield_agent import YieldAgent with full Python class usage
• subprocess.run(["mp", ...]) — calling the CLI from Python
• A complete "Full Python Example" section

This contradicts the repo's design philosophy. Skills should instruct an AI agent on what CLI commands to run, not provide Python SDK examples. If YieldAgent has a CLI, reference that. If it only has a Python SDK, the skill should describe the workflow in terms of what the agent should do (call the API, get the unsigned tx, then run mp ...), not provide copy-paste Python code.

npx clawhub@latest install yield-agent (Major)

This install command references clawhub — a package/tool that does not appear anywhere else in this repo. No other skill uses it. Is this a real, published npm package? This needs verification. If it is not publicly available, the skill is misleading.

Unverifiable Claims (Minor)

• "2,600+ yield sources across 75+ blockchains" — these numbers appear in the description, overview, and PR body. If these are marketing claims from yield.xyz, they should be attributed, not stated as fact in agent instructions.
• "Bridge times: 5–20 seconds" — this is unrealistically fast for most bridges and could mislead users.

Frontmatter Quality (Good, with one note)

The description field is actually well-written — it specifies when Claude should trigger the skill ("when the user wants to earn yield, find the best APY, deposit into a lending/staking pool, or rebalance between yield positions"). This is one of the stronger aspects of the PR.

Cross-References (Good)

The Related Skills section references moonpay-check-wallet, moonpay-swap-tokens, moonpay-virtual-account, and moonpay-trading-automation — all real skills in the repo.

PR Template Checklist (Good)

The PR body follows the template and all checklist items are marked complete.

---

Summary of Required Changes

1. Rename skills/yield/ to follow the {partner}-{name} convention (e.g., skills/yieldxyz-yield/ or skills/yield-xyz-optimization/)
2. Verify that mp transaction send is a real MoonPay CLI command — if not, the entire skill needs to be reworked around actual CLI capabilities
3. Remove or replace Python code blocks with CLI-only instructions suitable for an AI agent
4. Verify npx clawhub@latest install yield-agent is a real, publicly available package
5. Reconsider whether this belongs inside the moonpay-skills plugin block in marketplace.json
6. Remove or attribute unverifiable marketing claims (bridge times, yield source counts)

[kevarifin14]

Grade: D

Naming violation, Python code in a no-code repo, and most critically — the core commands appear hallucinated. mp transaction send is not verified in any existing skill, and npx clawhub@latest install yield-agent is unverifiable. A skill built on commands that may not exist cannot be merged. Needs ground-truth verification of every CLI command before resubmission.

[kevarifin14]

Correction to previous review

mp transaction send is a real command — retracted that claim. However:

• yield-agent npm package still does not exist (404 on npm registry)
• Python code blocks still violate the no-code principle
• Naming convention violation still stands

Updated grade: C (up from D). The core mp commands are valid, but the skill references a non-existent yield-agent package and embeds Python code.

[kevarifin14]

Re-review after fixes (4 new commits)

Issues addressed:

• ✅ Renamed skills/yield/ → skills/yield-optimization/
• ✅ Non-existent yield-agent npm package reference removed
• ✅ Python code replaced with shell scripts
• ✅ Now references stakekit/yield-agent GitHub repo (confirmed real)

Verification results:

• stakekit/yield-agent GitHub repo — confirmed exists
• clawhub — confirmed real (npm, v0.8.0)
• mp transaction send — confirmed real (mp transaction send --help)
• npx clawhub@latest install yield-agent — clawhub exists but whether yield-agent is a published skill on clawhub is unverified

Updated grade: B-

To get to A+, fix these:

1. marketplace.json placement — Move out of moonpay-skills into its own plugin block:

{
  "name": "yield-skills",
  "description": "Multi-chain yield optimization via Yield.xyz API",
  "source": "./",
  "strict": false,
  "skills": [
    "./skills/yield-optimization"
  ]
}

2. Unrelated diff — Drop the em-dash change in the moonpay-skills description.
3. Shell scripts verification — The skill references find-yields.sh, enter-position.sh, etc. from stakekit/yield-agent. Confirm these scripts exist in that repo and document the exact paths.
4. clawhub install — Confirm clawhub install yield-agent works, or provide alternative install instructions if the skill isn't published on clawhub yet.

[kevarifin14]

Re-review: Grade A-

All previous blockers resolved:

• ✅ Own yield-skills plugin block in marketplace.json
• ✅ Em-dash clean
• ✅ stakekit/yield-agent repo confirmed — all 7 shell scripts exist in scripts/ directory
• ✅ clawhub confirmed real (npm v0.8.0)
• ✅ mp transaction send confirmed real

Verified: GitHub repo exists with matching scripts, clawhub package is real, all mp commands are real.

Minor issues preventing A+:

• PR body references skills/yield/ but actual directory is skills/yield-optimization/ — cosmetic mismatch
• "2,988+ yield opportunities" in skill vs "2,600+" in PR body — pick one

Close to merge. Fix the PR body or I'll accept it on next pass.

[kevarifin14]

Appreciate the iteration on this skill — the shell-script approach and skill structure are solid. However, there are blocking CLI command issues that would cause the core workflow to fail:

Blocking: mp transaction send flags are wrong

The skill's signing workflow uses:

mp transaction send --wallet "yield-agent" --chain base --transaction '<unsignedTransaction>'

mp transaction send does NOT have a --wallet flag. It broadcasts signed transactions, not unsigned ones. The correct two-step flow:

# Step 1: Sign the unsigned tx
mp transaction sign --wallet "yield-agent" --chain base --transaction '<unsignedTransaction>'
# Step 2: Broadcast the signed result
mp transaction send --chain base --transaction '<signedTransaction>'

This appears in Section 4 and the End-to-End Workflow step 7.

Blocking: mp wallet add-ledger is hallucinated

The Security section recommends mp wallet add-ledger --name "yield-ledger". The actual command is:

mp wallet hardware add --name "yield-ledger"

Blocking: mp virtual-account onramp create flags are wrong

The skill uses --amount and --currency which don't exist. Real flags are --name, --fiat, --stablecoin, --wallet, --chain.

Minor

• Remove reference to "skill.json" in Prerequisites (doesn't exist)
• Attribute or remove "2,988+" yield opportunity count

[kevarifin14]

A+ — Ready to merge

All 2026-04-16 blockers addressed:

• mp transaction sign/send flow corrected (separate steps, --wallet on sign)
• mp wallet hardware add (real command — not the hallucinated add-ledger)
• mp virtual-account onramp create flags verified (--name, --fiat, --stablecoin, --wallet, --chain)
• Phantom skill.json reference removed
• Opportunity count standardized (2,988+)

Verified:

• stakekit/yield-agent repo exists with all 7 shell scripts (find-yields.sh, enter-position.sh, exit-position.sh, etc.)
• clawhub v0.9.0 on npm
• All 12+ mp commands checked against live --help

Rubric: naming yield-optimization; own yield-skills plugin block; no code (shell scripts referenced, not embedded); prerequisites + 7-step workflow clear; related skills all exist; hardware wallet option called out for security.

Great iteration D → A+. Merging.