HyperGrok Hyperliquid Setup

Prepare the desk computer to work with Hyperliquid - install the SDKs, pick testnet or mainnet, verify connectivity, and (only when the user asks) provision a trade-only API wallet…

Galleon Labs

@galleonlabs

Install

$ openclaw skills install @galleonlabs/hypergrok-hyperliquid-setup

Hyperliquid setup

Sections 1-3 are read-only and safe to run at any time. Section 4 provisions a key and is done only when the user asks to move to a testnet or mainnet desk. Section 5 is the readiness check the Execution Trader runs before its first send.

1. Networks

MainnetTestnet
RESThttps://api.hyperliquid.xyzhttps://api.hyperliquid-testnet.xyz
WebSocketwss://api.hyperliquid.xyz/wswss://api.hyperliquid-testnet.xyz/ws
Apphttps://app.hyperliquid.xyzhttps://app.hyperliquid-testnet.xyz
Faucet-https://app.hyperliquid-testnet.xyz/drip

Reads (/info) need no key. Writes (/exchange) need a signing key and are signed differently per network, so a key and its network go together. The desk records the network in /workspace/trading-desk/desk.md and in the environment.

Testnet is the default. Nothing on the desk selects mainnet unless HYPERLIQUID_NETWORK=mainnet is set on purpose.

2. Install

Python (used by most snippets on the desk):

python3 --version                                  # 3.9+ required, 3.11+ preferred
python3 -m pip install --user --upgrade "hyperliquid-python-sdk>=0.24,<1" \
  || python3 -m pip install --break-system-packages --upgrade "hyperliquid-python-sdk>=0.24,<1"
python3 -c "import hyperliquid, eth_account; print('ok', hyperliquid.__name__)"

Optional TypeScript (if the user prefers TS or wants the WebSocket client from @nktkas/hyperliquid):

node --version                                      # 22.12+ required by @nktkas/hyperliquid (ESM-only)
mkdir -p /workspace/trading-desk/ts && cd /workspace/trading-desk/ts
[ -f package.json ] || npm init -y >/dev/null
npm pkg set type=module >/dev/null                  # so .js files can use import and top-level await
npm i @nktkas/hyperliquid viem
node -e "import('@nktkas/hyperliquid').then(m=>console.log('ok', Object.keys(m).length))"

curl and jq are enough for every read in hyperliquid-market-data and hyperliquid-account; install jq if missing.

3. Connectivity check (no key)

for base in https://api.hyperliquid.xyz https://api.hyperliquid-testnet.xyz; do
  printf '%s -> ' "$base"
  curl -sS -m 10 -X POST "$base/info" -H 'Content-Type: application/json' \
    -d '{"type":"allMids"}' | jq -r 'if type=="object" then "ok, \(length) mids, BTC \(.BTC)" else "unexpected: \(.)" end'
done

Both lines should read ok. Record the time and result in the desk record. If one network fails, the desk is not blind on the other; say which.

4. API wallet (only when the user asks to trade)

An API wallet (Hyperliquid also calls it an agent wallet) is a separate key the user authorises to sign trading actions for their account. It can place, modify and cancel orders and change leverage and margin; it cannot withdraw to Arbitrum, cannot send USDC or tokens to another wallet, and cannot approve other agents or builders (those need the main wallet's signature). It can still move funds inside the user's own account (perp to spot and across dexs via agentSendAsset, into sub-accounts and vaults, and it can spend USDC on reserveRequestWeight), all of which the desk forbids by rule. That is the only kind of key that ever reaches the desk computer.

Why: all of the user's Bots share one computer, so anything on it is readable by every Bot. A trade-only key bounds the damage; a seed phrase or main-wallet key would not.

4.1 The user creates it in the Hyperliquid app

Guide the user; do not do it for them (it involves their wallet):

  1. Open the app for the chosen network (https://app.hyperliquid.xyz or https://app.hyperliquid-testnet.xyz), connect the main wallet, open the API page (/API, under More in the app's navigation).
  2. Generate a new API wallet, give it a name such as hypergrok-desk, choose a validity period (a few months is sensible; it can be revoked at any time), and Authorize. The app shows the API wallet's private key once. The API wallet's address is shown next to it.
  3. On testnet, fund the account from the faucet (/drip): it gives 1,000 mock USDC, but only to an address that has deposited on mainnet at some point (email/Privy logins get different addresses per network; export the mainnet wallet into a wallet extension and connect that to testnet). Testnet USDC is not real.

Never ask for the seed phrase or the main wallet's key. If the user offers it, decline and point back to this step.

4.2 The user provides the key through the secure secret store

Grok Bot has a secure secret card for exactly this. Ask the user to add a secret named HYPERLIQUID_PRIVATE_KEY with the API wallet's private key. Never ask them to paste it in chat and never write it to a file under /workspace.

If the desk's Grok Bot setup exposes secrets to the computer as environment variables, scripts read HYPERLIQUID_PRIVATE_KEY from the environment. If it does not, the fallback is a file the user creates while in control of the computer:

mkdir -p ~/.hyperliquid && chmod 700 ~/.hyperliquid
# the user pastes the key into this file themselves, then:
chmod 600 ~/.hyperliquid/api-wallet.key

Every signing snippet on the desk uses the same loader, which reads the environment first, then that file, and never prints the key:

import os
def load_key():
    k = os.environ.get("HYPERLIQUID_PRIVATE_KEY")
    if not k:
        p = os.path.expanduser("~/.hyperliquid/api-wallet.key")
        if os.path.exists(p):
            k = open(p).read().strip()
    if not k:
        raise SystemExit("no API wallet key available - see hyperliquid-setup section 4")
    return k

4.3 Non-secret configuration

Set these for the desk computer's shell (for example in ~/.bashrc, and mirror them in desk.md):

export HYPERLIQUID_NETWORK=testnet                 # or mainnet, deliberately
export HYPERLIQUID_ACCOUNT_ADDRESS=0xYourMainAccount   # the account the API wallet acts for, NOT the API wallet's address

The account address is public. Reads use it; the Execution Trader passes it as account_address when signing with the API wallet.

5. Readiness check (before the first send)

Run by the Execution Trader; read-only; proves the key belongs to the account on this network without sending anything.

import os
import eth_account
from hyperliquid.info import Info
from hyperliquid.utils import constants

def load_key():
    k = os.environ.get("HYPERLIQUID_PRIVATE_KEY")
    if not k:
        p = os.path.expanduser("~/.hyperliquid/api-wallet.key")
        if os.path.exists(p):
            k = open(p).read().strip()
    if not k:
        raise SystemExit("no API wallet key available - see hyperliquid-setup section 4")
    return k

network = os.environ.get("HYPERLIQUID_NETWORK", "testnet")
base = constants.MAINNET_API_URL if network == "mainnet" else constants.TESTNET_API_URL
account = os.environ["HYPERLIQUID_ACCOUNT_ADDRESS"]
agent_addr = eth_account.Account.from_key(load_key()).address   # derived locally, key never printed

info = Info(base, skip_ws=True)
role = info.user_role(agent_addr)                       # {"role": "agent", "data": {"user": "0x..."}} when approved
agents = info.extra_agents(account)                     # named agents: [{"address","name","validUntil"}, ...]
listed = any(a["address"].lower() == agent_addr.lower() for a in agents)
acts_for_account = role.get("role") == "agent" and role.get("data", {}).get("user", "").lower() == account.lower()
approved = acts_for_account or listed                   # an unnamed agent is valid but may not appear in extraAgents
state = info.user_state(account)

print(f"network={network} account={account}")
print(f"api_wallet={agent_addr} role={role.get('role')} approved_for_account={approved}")
for a in agents:
    if a["address"].lower() == agent_addr.lower():
        print(f"valid_until_ms={a.get('validUntil')} name={a.get('name')}")
print(f"account_value={state['marginSummary']['accountValue']} positions={len(state['assetPositions'])}")

Ready means: role=agent, approved_for_account=True, validUntil in the future (when the agent is named and listed), and account_value is what the user expects on this network. Anything else: stop and report which line disagrees.

6. Rotation and revocation

  • Revoke: the user removes the API wallet on the app's /API page. Sends from the desk fail immediately afterwards.
  • Rotate: revoke, create a new API wallet (4.1), replace the secret (4.2), rerun the readiness check. Journal the time.
  • Suspected misuse: revoke first, investigate second (desk-incident-response playbook G).

Pitfalls

  • Using the API wallet's address as HYPERLIQUID_ACCOUNT_ADDRESS. Reads on the agent's address show an empty account; the account is the main wallet's address.
  • Signing for mainnet with a testnet key or vice versa. The signature encodes the network; the exchange rejects the mismatch, but the desk should never get that far.
  • Installing the SDK into a temporary environment that does not survive a computer refresh. Prefer --user; re-run section 2 if imports fail after a restart.
  • Info(...) performs network calls on construction; pass skip_ws=True unless you want the WebSocket thread.
  • Printing environment variables while debugging. Redact anything that could be the key.

Top skills in this category

Skill Vetter

@spclaudehome

Security-first skill vetting for AI agents. Use before installing any skill from ClawdHub, GitHub, or other sources. Checks for red flags, permission scope, and suspicious patterns.

1.3k278k

Multi Search Engine

@gpyangyoujun

Multi search engine integration with 16 engines (7 CN + 9 Global). Supports advanced search operators, time filters, site search, privacy engines, and Wolfra...

764163k

Playwright (Automation + MCP + Scraper)

@ivangdavila

Automates, tests, and debugs browsers with Playwright: locators, auto-waiting, traces, CI runs, and MCP browser control. Use when a test is flaky, times out, or fails only in CI or headless; when a locator matches multiple elements or the wrong one (strict mode violation); when clicks need force, waits become sleeps, or networkidle never settles; for storageState and login setup, request mocking and HAR replay, uploads and downloads, iframes and shadow DOM, popups and dialogs, screenshot diffs that change per machine, trace and report artifacts, sharding a slow suite, device and permission emulation, accessibility checks, driving a real browser through Playwright MCP, extracting data from JS-rendered pages, or porting a Cypress, Puppeteer, or Selenium suite to Playwright. Not for maintaining an existing Cypress or Puppeteer suite (cypress, puppeteer) or for work a plain HTTP request answers (http).

13043k

Searxng

@abk234

Privacy-respecting metasearch using your local SearXNG instance. Search the web, images, news, and more without external API dependencies.

3223k

Skill Vetter - Pre-Install Security Review

@donovanpankratz-del

Security vetting protocol before installing any AI agent skill. Red flag detection for credential theft, obfuscated code, exfiltration. Risk classification L...

2726k