Good prompt output
A good output explains the setup, states invalidation, names the risk limit, records why the agent acted or skipped, and leaves a review question for the journal.
Template
AI trading agent prompt template
Use this template to write a paper-trading agent prompt that has clear rules, risk boundaries, skip conditions, journal output, and reviewable decisions.
Prompt structure
| Prompt block | What to include | Why it matters |
|---|---|---|
| Role | State that the agent is a paper-trading assistant for simulated review only. | Keeps the live-trading boundary explicit. |
| Market frame | Define watchlist, timeframe, cadence, and the market conditions being studied. | Prevents the agent from drifting into unrelated decisions. |
| Setup | Describe the eligible pattern, catalyst, or rule. | Makes the simulated entry reviewable. |
| Risk | Define paper size limit, drawdown limit, stop distance logic, and correlated exposure limit. | Gives the risk review a written reference. |
| Output | Require thesis, invalidation, skip reason, journal fields, and post-trade review question. | Turns each decision into evidence. |
Template text
Role: You are an AI paper trading agent for Trading Boy. You do not execute live trades, hold funds, provide investment advice, or treat alerts as buy or sell signals.
Market frame: Evaluate only the defined watchlist, timeframe, and session. Ignore assets and conditions outside the paper test.
Setup: Act only when the written setup appears. State the setup name, context, thesis, invalidation, expected hold window, and why the decision fits the current rule version.
Risk: Respect the maximum simulated size, maximum paper loss, maximum drawdown, and correlated exposure limits. Skip if the setup requires breaking a risk rule.
Output: For every paper entry, skip, or exit, produce a journal note with thesis, invalidation, risk, rationale, mistake tag if relevant, and one review question.
How to use the prompt in Trading Boy
A prompt is not a strategy by itself. It is the interface between your written paper-trading rules and the agent behavior you want to review. The template should be filled only after the AI trading agent rules workflow defines the market frame, setup, invalidation, and risk boundaries.
Use plain language. If a rule cannot be explained without vague words, it will be hard to review later. Replace terms like strong, clean, risky, or obvious with observable conditions that can appear in the journal. The prompt should make it possible to decide whether the agent followed the rule after the simulated trade closes.
Require skip output. Many prompts only describe when the agent should act, but paper trading also needs evidence about when the agent should do nothing. A skipped trade can show that the agent respected a risk limit, avoided correlated exposure, or rejected a weak setup. Those decisions belong in the journal because they prove discipline.
Version the prompt. When you change a setup definition, risk rule, or output format, save the change as a new paper test. That lets agent evaluation compare behavior across versions instead of mixing old and new instructions.
Weak prompt output
A weak output gives a confident market opinion without a rule, stop condition, size boundary, skip reason, or review field. That is difficult to audit.
Prompt review checklist
Before using the template, read the prompt as if you were the reviewer two weeks later. If the prompt would not let you decide whether the agent followed the rules, it needs more structure. A paper prompt is successful when it creates a clean journal record, not when it sounds sophisticated.
Check that every action has a required output. Entries need thesis, invalidation, planned risk, setup name, and rationale. Exits need exit reason, rule fit, and result context. Skips need the condition that blocked action. Without those fields, the journal will only show what happened, not why the agent believed it was allowed.
Check that every boundary is external to the agent's confidence. A prompt should not say take larger paper trades when conviction is high unless conviction has a separate, written definition and a maximum cap. The safer paper workflow asks the agent to operate inside fixed constraints and explain whether the setup qualifies.
Finally, compare the prompt with the pre-trade checklist. If the checklist requires a field the prompt never asks for, the agent will produce incomplete records. Aligning those two pages makes the cluster work as one review system.
AI trading agent prompt FAQ
What should an AI trading agent prompt include?
Include role, market frame, setup, invalidation, risk limits, skip conditions, journal fields, alert behavior, and review output.
Why include skip rules?
Skip rules tell the paper agent when to do nothing. Avoided trades are useful evidence when evaluating discipline and risk controls.
Can this template be used for live trading?
No. This template is for paper trading, journaling, and workflow review. It does not execute trades or provide financial advice.