galleonlabs DeFi derivatives
Translate a market or hedge request into explicit exposure, collateral requirements and venue-specific order evidence. Reuse official tooling; keep research, planning and authorized execution distinct.
Start with the requested task
Read task examples and user configuration for a completed example and recovery path. Apply only the workflow stages needed by the current request. Reuse supplied preferences and compatible tools; ask only for missing terms that affect this task. Public explanation or comparison does not require transaction fields. Keep any existing authorization within its exact scope.
Select the product and tools
Read providers for GMX, Drift, Derive, Jupiter and Boros, their official integration surfaces and version gaps. Read workflows for risk planning and order reconciliation. Evaluation cases provide bounded fixture scenarios.
For Hyperliquid, use the installed hyperliquid-analyze, hyperliquid-plan, hyperliquid-execute or hyperliquid-monitor skill when available. This pack provides the cross-venue comparison and handoff; it does not reproduce Hyperliquid signing or nonce rules. No other pack is required for the general workflow here.
Define the exposure
Identify network, protocol version, exact instrument/market, expiry, payoff denomination, collateral, margin mode, owner, subaccount and signer role. An ETH perpetual, an ETH option and an ETH funding-rate contract are different exposures. Do not normalize them into a single token price or label annualized funding as guaranteed yield.
Obtain fresh oracle/mark/index prices, order book or available capacity, current positions and orders, balances, margin requirements and accrued funding/borrow costs. State the observation time and source. Missing account data prevents a personalized risk or execution-ready claim; public market analysis can still proceed.
Build a bounded plan
Use the user's existing risk budget and approved terms. Specify size in both native venue units and economic notional, side, order type, limit/acceptable price, maximum slippage, collateral change, leverage/margin mode, expiry, reduce-only intent, fee recipients and maximum costs. Stress adverse price, funding, volatility and exit liquidity; calculate combined exposure across existing orders and positions, not just the new order. A stop order may trigger without filling.
For options include premium, contract multiplier, settlement asset, exercise/expiry model and uncovered short liability. For funding-rate hedges include settlement period, rate convention, maturity, collateral and basis mismatch against the exposure being hedged. A hedge across venues creates separate liquidation and settlement risks.
Authorize and reconcile
Use the official builder within the user's established execution boundary. A delegated key with limited withdrawals can still lose trading collateral. Granting or renewing delegation, changing margin mode, depositing and withdrawing are distinct financial actions; none is incidental setup. Never handle a key, create an agent wallet, insert a fee recipient or authorize a trade from this skill's presence.
Before manual changes to a managed account, read account control to reconcile active bots, open orders and an authorized handoff.
Follow the venue lifecycle: persist operation identity; separate prepared, submitted, created/resting, partially filled, executed, cancelled and unknown states; reconcile authoritative fills and account changes. Resolve unknown outcomes before retries. A keeper-created order, relayer acceptance or successful transport response is not a completed trade. Cancel and replace can race with fills; reread both before calculating remaining size.
Deliver
Return the market answer, comparison or proposed ticket with exact instrument and account identity, exposure, downside constraints, costs, source times, supported capabilities and material gaps. For monitoring, report position/order/margin changes with trigger conditions and a bounded next action. Creating a schedule, starting continuous trading or changing an existing automation requires the user's actual request; the skill does not start a background service.