Hyperliquid Advanced logo

Hyperliquid Advanced

Organization
galleonlabs
hyperliquid-advanced

Less common Hyperliquid actions and their rules - dead-man's switch (scheduleCancel), TWAP orders, spot orders, expiresAfter and nonces, API wallet approval from code, sub-account and vault addressing, HIP-3 dexs, and what the desk deliberately does not do (transfers, withdrawals, builder fees, staking). Write actions are Execution Trader only, on an approved ticket. Use when a ticket asks for one of these or when a user asks whether the desk can.

Overview

Publishergalleonlabs
Repositoryhypergrok-trading-desk
Skill namehyperliquid-advanced
Stars
145
Forks
31
Bundled files
1
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 1 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by galleonlabs on GitHub. Read the source before you install it.

Installation

Install the Hyperliquid Advanced AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/galleonlabs/hypergrok-trading-desk/tree/main/skills/hyperliquid-advanced
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/galleonlabs/hypergrok-trading-desk.git /tmp/hypergrok-trading-desk
mkdir -p .claude/skills
cp -r /tmp/hypergrok-trading-desk/skills/hyperliquid-advanced .claude/skills/hyperliquid-advanced
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Hyperliquid Advanced in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Hyperliquid Advanced on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Hyperliquid Advanced is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Hyperliquid advanced actions

Uses the same header as hyperliquid-orders (info, exchange, ACCOUNT, round_px, round_sz, new_cloid). Everything that signs is Execution Trader only, on a ticket.

Dead-man's switch (scheduleCancel)

Tells the exchange to cancel all of the account's open orders at a future time unless you push the time out or clear it. Useful when the desk runs a watch that could die while resting orders sit on the book. The user asks for it explicitly; the report states the time.

python
from hyperliquid.utils.signing import get_timestamp_ms
res = exchange.schedule_cancel(get_timestamp_ms() + 10 * 60 * 1000)   # 10 minutes out; must be >= 5 s in the future
res = exchange.schedule_cancel(None)                                    # clear it

TypeScript: await exchange.scheduleCancel({ time: Date.now() + 600_000 }); omit time to clear. Limits: at most 10 triggers per day per account (resets 00:00 UTC).

It cancels protective stops too, and it is not position-aware. If it fires while a position is open, that position is left naked on the exchange and nothing re-arms it automatically. So: it is for a desk with resting orders and no position, or for a user who has explicitly accepted that outcome for this account. Do not arm it as routine hygiene while a position is open. If it does fire with a position open, that is an unprotected position the moment it lands - declare it and run playbook D in desk-incident-response at priority, rather than treating re-arming as clean-up.

TWAP

Splits a size into slices at fixed intervals (30 seconds minimum) over m minutes: 5 minutes to 7 days per the docs, though the TypeScript SDK's schema caps m at 1440 (24 hours), so longer runs need the raw action. Minimum 100 USD total; each slice capped at 3% slippage; t: true randomises slice sizes by up to 20%. Not in the Python SDK's high-level Exchange; use TypeScript or the raw action.

ts
const twap = await exchange.twapOrder({ twap: { a, b: true, s: "10", r: false, m: 30, t: true } });
const twapId = (twap.response.data.status as { running: { twapId: number } }).running.twapId;
await exchange.twapCancel({ a, t: twapId });

Raw action: {"type":"twapOrder","twap":{"a":1,"b":true,"s":"10","r":false,"m":30,"t":true}}; response {"status":{"running":{"twapId":N}}} or {"status":{"error":"..."}}. Monitor with the twapStates / userTwapSliceFills WebSocket subscriptions or userTwapSliceFills reads; TWAP fills carry a zero hash. A TWAP is still one ticket; the ticket states size, minutes, randomisation and reduce-only.

Spot orders

Same order action; asset id is 10000 + index of the pair in spotMeta.universe; size decimals are the base token's szDecimals; price gets 8 - szDecimals decimals (still 5 significant figures); minimum order value is 10 quote tokens. The Python SDK resolves "PURR/USDC" (and app-style aliases such as "HYPE/USDC" where unambiguous) to the spot asset id for you:

python
# header from hyperliquid-orders, then spot-specific rounding from spotMeta
from hyperliquid.utils.signing import float_to_wire
pair_name = "PURR/USDC"
spot = info.spot_meta()
pair = next(p for p in spot["universe"] if p["name"] == pair_name)
base_dec = next(t for t in spot["tokens"] if t["index"] == pair["tokens"][0])["szDecimals"]

def round_px_spot(px, rounding=ROUND_HALF_UP):
    value = Decimal(str(px))
    if not value.is_finite() or value <= 0:
        raise ValueError("spot price must be finite and positive")
    max_dec = max(8 - base_dec, 0)
    exponent = min(0, max(value.adjusted() - 4, -max_dec))
    result = value.quantize(Decimal(1).scaleb(exponent), rounding=rounding)
    if result <= 0:
        raise ValueError("spot price rounds to zero")
    return result

def round_sz_spot(sz):
    value = Decimal(str(sz))
    if not value.is_finite() or value <= 0:
        raise ValueError("spot size must be finite and positive")
    result = value.quantize(Decimal(1).scaleb(-base_dec), rounding=ROUND_DOWN)
    if result <= 0:
        raise ValueError("spot size rounds to zero")
    return result

def bounded_px_spot(reference, slippage, is_buy):
    reference, slippage = Decimal(str(reference)), Decimal(str(slippage))
    if not reference.is_finite() or reference <= 0:
        raise ValueError("spot reference must be finite and positive")
    if not slippage.is_finite() or not 0 <= slippage < 1:
        raise ValueError("slippage must be a fraction in [0, 1)")
    bound = reference * (1 + slippage if is_buy else 1 - slippage)
    return round_px_spot(bound, rounding=ROUND_DOWN if is_buy else ROUND_UP)

sz, px = round_sz_spot("120"), round_px_spot("0.1234")
assert sz * px >= 10, "below the 10 quote-token minimum"
sdk_sz, sdk_px = float(sz), float(px)
assert Decimal(float_to_wire(sdk_sz)) == sz, "SDK changed the approved size"
assert Decimal(float_to_wire(sdk_px)) == px, "SDK changed the approved price"
res = exchange.order(pair_name, True, sdk_sz, sdk_px, {"limit": {"tif": "Gtc"}}, cloid=new_cloid())

These helpers retain Decimal values until the SDK boundary, preserve valid integer prices, and reject nonfinite or zero results. For a bounded spot IOC use bounded_px_spot with the ticket's reference and fractional slippage: buys round down within the ceiling and sells round up within the floor. Verify the final encoded SDK price against that bound before signing; never widen it to obtain a fill.

Read balances with spotClearinghouseState (hyperliquid-account). Spot has no leverage and no liquidation. Under unifiedAccount/portfolioMargin abstraction modes balances behave differently; the desk stays in the default mode unless the user changes it in the app.

expiresAfter and nonces

  • Every signed action carries a nonce (unix ms). The SDKs manage it. Nonces are per signer, so two processes signing with the same API wallet at the same time will collide; the desk keeps one Execution Trader and one process at a time.
  • expiresAfter (ms) makes an action void if it reaches the exchange after that time. Useful protection against a delayed duplicate after a reconnect: exchange.set_expires_after(get_timestamp_ms() + 60_000) in Python (applies to following L1 actions; must be None for user-signed actions), or per-call { expiresAfter } in TypeScript. A stale rejection costs 5x rate-limit weight, so keep it generous (a minute) rather than tight.
  • Hyperliquid retains the highest 100 nonces for each signer and permits out-of-order acceptance within that window. A newer nonce does not invalidate every lower nonce.
  • noop marks its own nonce used. To invalidate an unknown in-flight action, a recovery ticket must bind it to the original exact signed nonce, the same API wallet signer and the same network. Risk PASS, user approval by id and the native approval gate still apply; only the Execution Trader sends it, once. Confirm that the noop was accepted, then reconcile the original order and account. A nonce-reuse rejection does not confirm the noop landed or establish whether the original executed. If the original nonce is unavailable or the noop result is unconfirmed, keep the reservation and pause new risk; escalate to the Desk Lead and user. See desk-execution-protocol for expiry-based recovery.

Approving an API wallet from code

The desk's normal path is the app (hyperliquid-setup section 4), because approval must be signed by the main wallet and that key never touches the desk computer. For completeness: Exchange(main_wallet).approve_agent(name) in Python generates a fresh agent key and returns (result, agent_private_key); TypeScript exchange.approveAgent({ agentAddress, agentName }) approves an address you generated. Named agents can carry an expiry via "name valid_until <ms>" (up to 180 days); an account may hold 1 unnamed and up to 3 named agents, plus 2 named per sub-account; re-approving the same name (or a new unnamed agent) replaces the previous key. Never reuse a revoked agent address.

Sub-accounts and vaults

Orders can be sent for a sub-account or vault by setting vaultAddress (Python: Exchange(..., vault_address="0x..."); TypeScript: defaultVaultAddress or per-call { vaultAddress }); the API wallet of the master signs. Reads for that address use the sub-account/vault address as user. The desk supports this only if the user asks and records it in desk.md; it never creates sub-accounts or transfers funds into or out of them.

HIP-3 builder dexs

Other perp dexs exist beside the main one (perpDexs). Coins are dex:COIN, asset ids are 100000 + 10000 x dex index + index in that dex's meta, margin is often isolated-only, and reads take a dex parameter. Off by default on the desk; if the user wants one, everything above applies with the prefixed coin name and the dex's own meta.

Rate limits and reserveRequestWeight

Address-based limits for actions: a buffer of 10,000 plus 1 per 1 USDC of cumulative volume; when exhausted, 1 action per 10 seconds (cancels get extra headroom). Check userRateLimit. An account can buy more with reserveRequestWeight (0.0005 USDC each) - the desk does not do this automatically; mention it if the user hits the limit. Per-IP /exchange weight is 1 + floor(batch size / 40).

Deliberately not on this desk

These exist in the API and SDKs; the desk does not use them, and the user does them in the Hyperliquid app with their main wallet:

ActionWhat it does
usdSend, spotSend, sendAssetsend USDC or tokens to another address or between dexs (agentSendAsset, the self-only variant, is L1-signed and agent-capable)
withdraw3withdraw USDC to Arbitrum
usdClassTransfermove USDC between perp and spot balances
vaultTransfer, subAccountTransfer, createSubAccountmove funds into or out of vaults and sub-accounts (L1-signed, so an API wallet technically can; the desk's rule is the guard)
approveBuilderFee, builder on orderslet a builder charge a fee on your orders
cDeposit, cWithdraw, tokenDelegateHYPE staking
userSetAbstraction and friendschange the account's margin mode

If a ticket asks for one of these, the Execution Trader declines and the Desk Lead explains why (desk-operating-model, "Excluded on purpose").

Pitfalls

  • A dead-man's switch that fires also removes stops. Re-arm protection.
  • TWAP minimum size and duration errors (Invalid TWAP duration) come back inside status, not as resting.
  • Spot size decimals come from the base token, not the pair; ids differ per network.
  • Reusing an agent address after revocation can replay old signed actions; generate fresh keys.
  • Two signers on one API wallet in parallel: nonce collisions and rejected actions.

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Hyperliquid Advanced AI skill do?

Less common Hyperliquid actions and their rules - dead-man's switch (scheduleCancel), TWAP orders, spot orders, expiresAfter and nonces, API wallet approval from code, sub-account and vault addressing, HIP-3 dexs, and what the desk deliberately does not do (transfers, withdrawals, builder fees, staking). Write actions are Execution Trader only, on an approved ticket. Use when a ticket asks for one of these or when a user asks whether the desk can.

Why use Hyperliquid Advanced on TypingMind?

Because you install it once and use it with any model. Hyperliquid Advanced is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Hyperliquid Advanced in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/galleonlabs/hypergrok-trading-desk/tree/main/skills/hyperliquid-advanced. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Hyperliquid Advanced?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Hyperliquid Advanced?

As many as you like. As long as a model supports skills, you can use Hyperliquid Advanced with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Hyperliquid Advanced AI skill free?

Yes. It is published on GitHub by galleonlabs under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇