Define the scope
Write the market universe, watchlist, timeframe, setup, and cadence before the agent records any simulated decision.
Workflow
AI trading agent rules workflow
Use this workflow to turn an AI trading agent idea into explicit paper-trading rules that can be tested, journaled, and reviewed before any live-capital workflow is considered.
Define the boundary
Write invalidation, skip rules, size limits, maximum paper drawdown, and conditions that block new entries.
Define the evidence
Write the journal fields and review questions that will prove whether the paper agent followed the rules.
Rule layers for an AI paper trading agent
| Rule layer | What to write | Review evidence |
|---|---|---|
| Market frame | Assets, timeframe, session, and conditions where the agent is allowed to evaluate. | Journal entries show the agent stayed inside the intended universe. |
| Setup rule | The pattern, catalyst, or condition that makes a paper trade eligible. | Decision rationale explains why the setup matched the written rule. |
| Invalidation rule | The price, condition, or time limit that proves the paper thesis wrong. | Post-trade review can compare the exit with the original invalidation. |
| Risk rule | Maximum simulated size, loss, drawdown, and correlated exposure. | Risk review confirms the agent did not exceed the limit. |
| Skip rule | Conditions where the agent must do nothing, even if part of the setup appears. | Skipped decisions become useful evidence instead of missing data. |
How to write rules that can be reviewed
A paper-trading rule is only useful if another person can read it later and decide whether the agent followed it. Rules like find strong trades or avoid bad setups are too vague. They sound useful during setup, but they do not create clean review evidence after the simulated trade closes.
Write rules in observable language. Instead of telling an agent to look for momentum, define the conditions that make momentum eligible for the workflow. Instead of telling the agent to manage risk carefully, define maximum paper size, maximum drawdown, and skip conditions. The goal is not to make the agent more confident. The goal is to make the agent easier to audit.
Keep each rule version stable long enough to collect a meaningful paper sample. If the watchlist, setup, risk limit, and exit rule all change at the same time, the next review will not know which change mattered. Trading Boy works best when a rule change has a version label and a reason that appears in the journal.
Connect the rules to the rest of the cluster. Use the AI trading agent prompt template to turn the rule into structured instructions, the AI agent risk controls workflow to set boundaries, and AI paper trading agent evaluation to score the sample.
Example rule set
Agent frame: The paper agent watches a defined crypto watchlist on one timeframe and only evaluates trend-continuation setups after the review window starts. It must ignore assets outside the watchlist.
Entry rule: A simulated entry is eligible only when the setup matches the written condition, the invalidation level is known, and planned paper size stays under the risk limit.
Skip rule: The agent must skip if the target is unrealistic, if similar exposure is already open, if the invalidation is missing, or if the setup appears only after the move is extended.
Review rule: Every decision must create a journal note with setup, invalidation, risk, decision rationale, and one question for the post-trade review.
When to change a rule
Change a rule when repeated paper evidence shows the agent is too early, too late, too frequent, oversized, or unclear. Do not change multiple rules from one isolated result.
When to keep collecting data
Keep the version stable when the agent follows the rule but the sample is still too small. More paper evidence may be better than immediate optimization.
Common rule-writing mistakes
The first mistake is writing a rule that only describes what a good trade looks like after it works. A useful rule must also describe when the agent should skip, pause, or mark the setup as incomplete. Those negative conditions are what make paper trading reviewable.
The second mistake is mixing strategy language with review language. Strategy language says what the agent is looking for. Review language says how the decision will be judged later. A complete paper-agent rule set needs both, because the journal has to compare the original intent with the actual simulated behavior.
The third mistake is letting the agent decide its own boundaries. Keep market frame, size, drawdown, and skip conditions outside the agent's discretion. The agent can explain whether a condition is met, but the paper workflow should define the condition first.
AI trading agent rules FAQ
What rules should an AI paper trading agent have?
It should have rules for market universe, setup, invalidation, risk size, skip conditions, alert behavior, journal fields, and review cadence.
Why should agent rules be versioned?
Versioning ties a paper sample to the exact instructions that created it, which helps compare behavior before and after a focused change.
Can clear rules make an agent safe for live trading?
No. Clear rules improve paper-trading review quality, but they do not remove market risk, guarantee future performance, or provide financial advice.