Skip to content

Back to the catalog

galleonlabs/hypergrok-trading-desk
github.comrepo

galleonlabs/hypergrok-trading-desk

HyperGrok Trading Desk is an open-source "Agent Skills" pack (MIT licensed) that turns a Grok Bot, Claude Code or Cursor instance into a coordinated seven-agent trading desk for Hyperliquid perpetuals, using seventeen portable SKILL.md skills covering bootstrap/setup, exchange API access, and desk operating procedures. Each trade moves through a fixed pipeline — idea, evidence, risk sign-off, explicit human approval by ticket id, a single execution agent that sends the order, reconciliation, and a separate review agent — with read-only "opening bell" and "desk doctor" scripts for zero-key market snapshots and release/connectivity checks. The design choices worth noting for agent builders are the one-writer/many-readers pattern, sizing risk off the stressed (slipped) cost of a stop rather than its trigger price, and keeping the approval gate (Grok Bot's Require Approval rule) outside the chat the agent itself can write to.

TL;DR A 7-subagent, 17-skill SKILL.md pack that turns Grok Bot/Claude Code/Cursor into an approval-gated Hyperliquid trading desk.
Document

What it claims

A pinned release (v1.4.8) of seventeen SKILL.md skills and seven role prompts that build a Hyperliquid perps desk across Grok Bot, Claude Code, Cursor or Grok Build. It starts in research mode using only Hyperliquid's public /info endpoint, with no wallet, account read or order. You add a testnet API wallet for paper trading, then a mainnet one.

How it works

Six Bots share a "Trading Floor" group chat. The Trade Reviewer works by DM.

Approval gate. Approval is typed as approve <ticket-id>. That line is only evidence. Enforcement comes from Grok Bot's Require Approval rule, which sits outside the chat.

flowchart LR
  A[Risk ticket posted] --> B{Require Approval rule<br/>outside chat}
  B -- approved by ticket id --> C[Execution Trader: one grouped send]
  B -- not approved --> D[No order]
  C --> E[Reconcile by client order id]
  E --> F[Reviewer DM: process vs outcome]

Sizing arithmetic. The README shows a worked example, which you can check:

  • Equity is $10,200, read from clearinghouseState. Risk is 0.5%, which is $51.
  • The entry is 3,000 and the stop is 2,900, so R = 100.
  • Stressed cost per ETH is 100 + 3.00 slippage + 2.65 fees = 105.65.
  • Size is 51 / 105.65 = 0.4827 ETH (about $1,448).
  • The entry is sent as a limit Gtc order. The stop is a market sell sent in the same grouped order.

Other controls:

  • Key scope. The only key is a trade-only API wallet held in Grok Bot's secret store. It can trade but cannot withdraw.
  • Limits. Your limits file can only tighten the desk's built-in caps, never loosen them.
  • Testnet first. Every new kind of action is rehearsed on testnet before mainnet.

Opening Bell. opening_bell.py reports:

  • mid, mark and oracle prices
  • 24h change and volume
  • hourly funding and open interest
  • spread and depth at 5/10/25 bps

Twenty book levels do not reach the outer bands on a liquid perp. The script therefore re-reads the book at a coarser nSigFigs page, labels which page each figure came from, and marks unreached bands as floors rather than totals.

Strategist. It runs backtests that include fees and an out-of-sample split. It reports the trade distribution rather than a single headline number. Strategies can then be paper-traded on testnet.

Monitoring. WebSocket routines watch the feed and alert you, for example when funding flips negative.

Skill coverage. The Hyperliquid skills cover:

  • limit, IOC and TP/SL orders, including grouping and client order ids
  • positions and margin
  • WebSockets
  • dead-man's switch, TWAP and spot

Reads use curl. Signing uses the official Python SDK or @nktkas/hyperliquid.

How to use it

Grok Bot: add "HyperGrok Desk Lead" and send Start the desk. Alternatively, paste a prompt pointing at skills/hypergrok-bootstrap/SKILL.md and SETUP.md at tag v1.4.8. The bootstrap ends with a receipt that separates what was done from what still needs a manual step.

Read-only checks (standard-library Python):

python3 scripts/opening_bell.py --coin ETH

scripts/desk_doctor.py checks the release, team files, reviewed skill bytes, desk folders and public connectivity.

Claude Code:

/plugin marketplace add galleonlabs/hypergrok-trading-desk
/plugin install hypergrok@hypergrok

Grok Build (tested with 1.0.30):

grok plugin install galleonlabs/hypergrok-trading-desk@v1.4.8 --trust

or

grok plugin marketplace add galleonlabs/hypergrok-trading-desk
grok plugin install hypergrok --trust

As a skill pack:

npm exec --package=skills@1.5.23 -- skills add galleonlabs/hypergrok-trading-desk

Cursor: open the repo and enable the plugin. On every platform, run /desk-operating-model to begin.

template/ holds the exact public Grok Bot profile and skill hashes, which you can use to verify what you installed.

Limits

  • Grok Build rules. Grok Build loads skills/ and agents/, but loading of rules/ is not verified. Rules-based guardrails may not apply there.
  • Approval outside Grok Bot. The external approval gate is Grok Bot's Require Approval rule. The page does not say what enforces approval in Claude Code or Cursor.
  • Liquidation risk. The page warns that perpetuals can liquidate an account. It describes itself as documentation and instructions, not financial advice.
  • Installing grants nothing. Installing the skills does not connect a wallet or authorise a trade.
  • Licence. MIT. The copyright and permission notice must be retained when reusing copies or substantial portions.

Organisations named here

  • Galleon Labs
  • Hyperliquid
  • pump.fun
  • Andrew Wilkinson
  • Claude Code
  • Cursor
  • Grok

Technologies named here

  • Grok Bot
  • HyperGrok Trading Desk
  • Hyperliquid
  • Claude Code
  • Cursor
  • Grok Build
  • skills
  • Hyperliquid official Python SDK
  • @nktkas/hyperliquid
  • opening_bell.py
  • desk_doctor.py
  • Boomkin
  • crypto-defi-skills
  • SKILL.md format
  • Grok
  • Git
  • Python

Techniques named here

  • Multi-agent trading desk architecture
  • Trade lifecycle protocol
  • Sizing on a stressed stop
  • One writer, six readers
  • Approval gate enforced outside the chat
  • Progressive coarsening of order-book reads