Build on HypurrFi with AI Agents: Claude Code, Cursor, and Codex

If you build on HypurrFi with AI agents, the hardest part is context. Contract addresses, market types, oracle logic, USDXL minting paths, EVC batching, the difference between a Pooled Market and a HypurrFi Markets vault. That context lives in one file: SKILL.md, hosted as a public static asset on the HypurrFi docs site. Point your agent at it and the agent writes integration code that actually runs on HyperEVM.

HypurrFi maintains dedicated agent documentation for Claude Code, Cursor, and OpenAI Codex-style workflows. The docs are structured for retrieval: code-first, flat, self-contained, with all addresses inline. That is the difference between "the LLM hallucinates a wrong Pool address" and "the agent ships a working supply flow in one shot." If you are new to the broader Hyperliquid stack, start with DeFi lending on Hyperliquid for context on what the agents will actually be wiring together.

Key takeaways

• HypurrFi publishes agent-specific docs with contract addresses, code patterns, and product taxonomy in a format LLMs can read directly.

• One file, SKILL.md, carries full protocol context. Drop it into Claude Code, Cursor, or Codex and the agent knows Prime, Yield, Scale, Earn Vaults, Pooled Markets, USDXL, and the HypurrFi Crypto Credit Card without extra prompts.

• The AGENTS.md standard (agents.md) is how HypurrFi keeps Claude Code, Cursor, OpenAI Codex, OpenClaw, and Hermes on the same instructions.

• Dedicated setup pages exist for Claude Code and Cursor. Codex, OpenClaw, and Hermes all follow the same AGENTS.md convention and work without extra configuration.

• HypurrFi secured over $350M in peak exposure on audited Euler v2 and Aave v3 architectures (DeFiLlama). Your agent should treat these as production contracts, not testbeds.

What is a HypurrFi AI agent workflow?

A HypurrFi AI agent workflow is a setup where a coding agent (Claude Code, Cursor, or OpenAI Codex) loads HypurrFi's SKILL.md and AGENTS.md as primary context, then writes, reviews, or edits integration code targeting HypurrFi's lending markets on Hyperliquid. The agent inherits accurate addresses, product names, and code patterns without needing to scrape the docs site.

This matters because DeFi integration code fails silently on wrong addresses, wrong decimals, or wrong interest rate modes. A Pool address that is one character off will compile, deploy, and revert every call. Giving the agent a curated reference file shifts those errors from "agent hallucinated" to "agent followed the canonical spec."

Why does HypurrFi ship an AGENTS.md and SKILL.md?

Two reasons: the AGENTS.md standard is the common ground across agent tools, and SKILL.md is the deep context that makes protocol-specific code correct.

AGENTS.md is an open, agent-agnostic convention (used by OpenAI Codex, Cursor, Claude Code, OpenClaw, Hermes, Aider, Warp, and others) that sits in a repo root and tells any agent how to work in that codebase: commands, terminology, style rules, constraints. It is effectively the README.md an AI reads before touching a keyboard.

SKILL.md goes further. It carries the HypurrFi protocol reference: product taxonomy (Prime, Yield, Scale, Earn Vaults, Pooled, USDXL), contract addresses (Pool, PoolAddressesProvider, eVaultFactory, EVC, HyFiOracle), and reference code patterns for supply, borrow, repay, withdraw, and EVC batching. It is the thing an agent needs to actually write working HyperEVM integration code.

Both files are served as public static assets on the docs site, so any agent can fetch them directly without a GitHub account:

• docs.hypurrfi.com/AGENTS.md

• docs.hypurrfi.com/SKILL.md

• docs.hypurrfi.com/llms.txt and docs.hypurrfi.com/llms-full.txt for LLM-index consumers

Which agent should you use with HypurrFi?

All five work with HypurrFi today. Pick the one that fits your workflow. The context files are the same either way.

How do you set up Claude Code for HypurrFi?

Claude Code picks up AGENTS.md and SKILL.md automatically when they sit in the project root. Full setup lives at docs.hypurrfi.com/ai-tools/claude-code. The short version:

1. Install Claude Code with npm install -g @anthropic-ai/claude-code.

2. Fetch SKILL.md and AGENTS.md from the docs site and drop them into your integration repo root: curl -O https://docs.hypurrfi.com/SKILL.md && curl -O https://docs.hypurrfi.com/AGENTS.md.

3. Run claude in the project root. The agent reads AGENTS.md and SKILL.md on startup.

4. Ask it to integrate. Example prompt: "Write a TypeScript function that supplies USDT0 to HypurrFi Pooled Markets and borrows USDXL against it using viem."

Claude Code will pull addresses, ABIs, and patterns directly from SKILL.md without you pasting them into chat.

How do you set up Cursor for HypurrFi?

Cursor ships with AGENTS.md support and respects .cursor/rules/ directories. See docs.hypurrfi.com/ai-tools/cursor for the canonical setup. Steps:

1. Open your integration repo (or the HypurrFi docs repo) in Cursor.

2. Attach SKILL.md to the agent context using the @ file reference or by pinning it in the sidebar.

3. Confirm Cursor is reading AGENTS.md from the root. You should see it listed under the project rules.

4. Start editing. Ask Cursor to generate a borrow-and-repay helper or to audit an existing integration against HypurrFi's current addresses.

Cursor's inline diff flow pairs well with HypurrFi's flat agent docs because the agent can cite specific addresses and patterns from SKILL.md as it proposes edits.

How do you set up Codex for HypurrFi?

OpenAI Codex (the CLI agent and the IDE extensions) follows the AGENTS.md standard. There is no HypurrFi-specific plugin required; the convention is enough.

1. Install the Codex CLI or enable the Codex IDE extension.

2. Make sure AGENTS.md is in the project root. HypurrFi's AGENTS.md covers terminology, style rules, and product taxonomy.

3. Include SKILL.md in the same root. Reference it explicitly in your first prompt: "Read SKILL.md before writing any integration code."

4. Prompt for integration work. Example: "Using the addresses in SKILL.md, write a Foundry test that deposits HYPE into HypurrFi Prime via the EVC batch call."

Codex-style agents do best when you anchor them on the source of truth in the first message. SKILL.md is that source of truth for HypurrFi.

What should you point your agent at in the docs?

HypurrFi's docs are split into human pages and agent pages. For integration work, send your agent directly to the agent pages. They are flat, code-first, and self-contained.

For developer-focused human docs (guides, smart contract APIs, addresses), see docs.hypurrfi.com/developers/overview.

Should you also load Euler's agent skills?

Yes, if you are writing integration code against HypurrFi Markets (Prime, Yield, or Scale), layer Euler's official agent-skills on top of HypurrFi's SKILL.md. The two skill bases are complementary: HypurrFi's file carries deployment-specific context (addresses, risk parameters, which vaults are live on HyperEVM), while Euler's skills carry the generic Euler v2 mechanics (vault operations, EVC batching patterns, oracle adapters, IRM types, EulerEarn).

Euler ships five installable skills via npx add-skill euler-xyz/agent-skills:

Install the core skill with:

Then drop docs.hypurrfi.com/SKILL.md into the same repo root. The agent gets generic Euler v2 vault expertise from the Euler skills and HypurrFi-specific deployment context from SKILL.md. This mirrors how Euler's own skill base is structured, so the conventions line up rather than conflict.

For Pooled Markets (Aave v3) work, SKILL.md plus the Pooled Markets agent reference covers the full surface. No additional third-party skill base needed.

What does a good HypurrFi agent prompt look like?

Anchor every prompt on three things: the product, the action, and the source of truth. HypurrFi's agent docs are the source of truth.

Weak prompt:

"Write code to borrow on HypurrFi."

Strong prompt:

"Using the Pool address and ABI from SKILL.md, write a TypeScript function with viem that supplies 1,000 USDT0 to HypurrFi Pooled Markets, then borrows 500 USDXL at variable rate mode against it. Reference docs.hypurrfi.com/agents/pooled-markets for the exact borrow signature. Do not invent addresses."

The strong prompt names the product (Pooled Markets), the action (supply then borrow), and the source of truth (SKILL.md plus the agent reference page). It also blocks the common failure mode ("do not invent addresses").

What integrations are agents good at today?

Agents handle the mechanical parts of HypurrFi integration well when they have SKILL.md loaded:

• Writing viem or ethers supply, borrow, repay, withdraw flows against Pooled Markets

• Generating EVC batch calls for HypurrFi Markets (Prime, Yield, Scale)

• Building USDXL mint and repay helpers using the Pooled Markets CDP path, including flows that let users borrow against crypto without selling

• Wiring up oracle price reads from HyFiOracle

• Writing Foundry or Hardhat tests against the audited Euler v2 and Aave v3 interfaces, especially when paired with Euler's agent-skills for deep EVC and vault semantics

• Drafting documentation copy and integration snippets that line up with the same context file the agent is reading

Risk modeling, liquidation simulation, and parameter tuning still want a human in the loop. Treat the agent as a fast, accurate first pass, not a final reviewer.

Frequently asked questions

Do I need a special API key from HypurrFi to use an AI agent?

No. HypurrFi does not gate agent access. You use your own Claude, Cursor, or OpenAI credentials. HypurrFi publishes AGENTS.md and SKILL.md as public static files on the docs site, and agent reference pages are live at docs.hypurrfi.com/agents.

Will an AI agent write correct contract addresses?

Only if you point it at SKILL.md or the agent reference pages. Without that grounding, LLMs hallucinate addresses. SKILL.md carries the current Pool, eVaultFactory, EVC, HyFiOracle, and other addresses inline. As long as the agent reads SKILL.md first, the addresses in generated code will match what is deployed on HyperEVM.

Can I use HypurrFi agent docs in OpenClaw or Hermes?

Yes. OpenClaw and Hermes both follow the AGENTS.md convention. For OpenClaw, put AGENTS.md and SKILL.md in your repo root, and the agent picks them up per workspace. For Hermes, keep the files in the project root or stage them under ~/.hermes/ so the agent loop reads them at startup. No HypurrFi-specific plugin is required for either.

How often does HypurrFi update SKILL.md?

SKILL.md is served as a public static file at docs.hypurrfi.com/SKILL.md. It updates alongside protocol changes (new markets, new assets, address changes, new products like the HypurrFi Crypto Credit Card or Scale). For the latest version, always fetch from the docs URL rather than a cached copy.

What is the difference between AGENTS.md and SKILL.md?

AGENTS.md tells the agent how to work in the repo: commands, terminology, style rules, compliance constraints. It is short and procedural. SKILL.md is the deep protocol reference: product taxonomy, contract addresses, code patterns. It is long and technical. Agents read both. Developers writing integration code care most about SKILL.md. Developers editing HypurrFi docs care most about AGENTS.md.

Should I use Euler's agent-skills alongside HypurrFi's SKILL.md?

Yes for HypurrFi Markets work (Prime, Yield, Scale). The two skill bases are complementary. Euler's agent-skills teach the agent generic Euler v2 mechanics (vault operations, EVC batching, oracle adapters, IRM types), while HypurrFi's SKILL.md provides deployment-specific context (which vaults are live on HyperEVM, current addresses, risk parameters). Install with npx add-skill euler-xyz/agent-skills --skill euler-vaults and keep SKILL.md in the repo root. For Pooled Markets (Aave v3) work, SKILL.md alone is sufficient.

Summary

HypurrFi publishes agent-ready docs so Claude Code, Cursor, OpenAI Codex, OpenClaw, and Hermes can write correct HyperEVM integration code on the first try. The canonical entry points are AGENTS.md (repo style and terminology), SKILL.md (full protocol reference with contract addresses and code patterns), and the dedicated agent pages at docs.hypurrfi.com/agents. Setup guides for Claude Code and Cursor cover installation and context loading. OpenAI Codex, OpenClaw, and Hermes use the same AGENTS.md standard with no extra configuration. HypurrFi is live on Hyperliquid, secured over $350M in peak exposure across audited Euler v2 and Aave v3 markets (DeFiLlama), and runs on the same codebases the rest of DeFi trusts. Point your agent at the docs and ship.

Last updated: April 2026

Sources

• HypurrFi Agent Overview

• HypurrFi SKILL.md

• HypurrFi AGENTS.md

• AGENTS.md standard

• Claude Code setup

• Cursor setup

• OpenClaw docs

• Hermes Agent docs

• Euler agent-skills

• DeFiLlama - HypurrFi