OpenPond
Login

How to Prompt OpenPond

April 16, 2026
5 min read
0xglu
AIprompting

Most prompt advice is too general for building trading agents.

OpenPond works best when your prompt makes the job clear first. Are you asking a market question, drafting a strategy, building an app, or copying a template? Once that is clear, the rest of the prompt can focus on the details that actually change the result.

Start with the job

If you want a strategy draft before anything is built, say that.

Draft a live trading strategy for BTC-USDC before building: enter when EMA 8 crosses above EMA 21 and exit when EMA 8 crosses below EMA 21. I have not chosen trade size yet.

If you want the agent to build, say create or build.

Create a BTC-USDC breakout bot: go long when price closes above the 20-period Donchian high and exit when price closes below the 10-period Donchian low. Use 1h candles and 100 USDC per trade.

If you want a market answer, ask the question directly.

What is the current BTC-USDC trend on the daily chart?

That first sentence matters. It tells the system whether to answer, plan, or start creating the app.

Name the strategy pieces

For trading strategies, a strong prompt usually includes five pieces:

  • Market: the asset or pair, like BTC-USDC.
  • Signal: the condition to watch, like RSI, EMA cross, Donchian breakout, funding, or a news event.
  • Action: what to do when the signal appears.
  • Exit: what closes or reduces the position.
  • Execution: timeframe, trade size, cadence, leverage, slippage, or rebalance rules.

For example:

Create a BTC-USDC strategy: enter long only when RSI(14) is below 30 and price closes below the lower Bollinger Band. Exit when RSI is above 55 or price closes above the Bollinger middle band. Use 1h candles and 100 USDC per trade.

This gives the system enough information to understand both the entry and the exit. It also avoids vague work like "make me a trading bot," which forces the agent to invent the strategy for you.

Say what should happen after the first trade

Many real strategies need state. They might wait before re-entering, add to a winner, take partial profit, or rebalance multiple legs.

Those details belong in the prompt.

Build a BTC-USDC strategy: after any exit, wait 6 candles before entering again. Entry is RSI below 30, exit is RSI above 55. Use 1h candles and 100 USDC per trade.

Build a BTC-USDC strategy: enter 50 USDC on EMA 8/21 bullish cross, add another 50 USDC if price moves 1% in favor, max two entries, exit all on bearish cross.

Create a BTC-USDC bot: enter long on RSI below 30, sell half at +2%, sell the rest when RSI goes above 60, stop out at -1%. Use 100 USDC.

These prompts are still short, but they describe the lifecycle of the trade instead of only the entry.

Use natural language for complex strategies

You do not need to write a formal spec. Natural language is fine if the important constraints are present.

Create a BTC delta-neutral carry strategy: only open when funding is at least 4 bps and basis is under 30 bps, target 500 USDC per leg, rebalance every 30 minutes.

Create a strategy that goes long ETH-USDC and short BTC-USDC when ETH/BTC breaks above its 30-day moving average, and reverses when it breaks below. Use 500 USDC per leg.

For baskets, triggers, and TWAPs, include the thresholds and sizing.

Create a trigger bot: if BTC crosses above 100000, buy 70% ETH and 30% SOL with 500 USDC total. If BTC falls below 95000, sell the basket.

Create a TWAP bot that buys 10,000 USDC of BTC over 4 hours in randomized clips, only on weekdays, with max slippage 20 bps.

Copy templates with exact URLs

If you already know which template to start from, paste the exact OpenPond URL and say what to change.

Copy the Hyperliquid delta-neutral strategy from https://openpond.ai/openpondai/hyperliquid-delta-neutral.git and set it to use BTC with $5 margin at 8x leverage, rebalance every 15 minutes, and close or rebalance if delta drifts more than 3%.

For a pure copy, you can keep it even shorter.

Copy my-openpond strategy from https://openpond.ai/openpondai/my-openpond

What good prompting means here

Good prompting in OpenPond is not about sounding polished. It is about being legible.

Say the job first. Then give the market, signal, action, exit, and execution details. If one of those details is unknown, say that too. The system can draft the strategy, ask for the missing detail, and then build when you are ready.