One Rule a Week: Using Claude Code or Codex to Turn Your Trading Review Into a Real Upgrade Loop

Why 'just review harder' never works

Scrolling screenshots cannot tell you what actually cost you money this week. You remember the trades that felt bad, not the ones that quietly bled. So nothing specific changes, and next week looks like last week.

Two things fix that: a consistent record of what you did, and a specific question asked against it. That is the whole reason to bring AI in, not to predict the market, but to read your own history back to you honestly.

The loop, one week at a time

The entire loop is five small steps on a fixed weekly rhythm. Nothing here requires code.

• Daily: log each trade with a fixed set of fields, setup, planned entry, actual entry, exit, and any rule you broke.
• Friday: feed the week's record to Claude Code or Codex with ONE fixed prompt: 'Here are my trades and my rules. Rank my three most expensive repeated mistakes by total cost.'
• Weekend: take the number-one mistake and write a single rule that would have prevented it.
• Monday: deploy that one rule and change nothing else about your strategy.
• Next Friday: measure. Did that mistake shrink? Keep the rule, adjust it, or retire it, then pick the next leak.

A concrete example

Say Friday's output comes back: 'Your biggest leak this week, you widened your stop on 4 of 6 losing trades, average extra loss 0.8R.'

Your rule writes itself: 'My stop is set at entry and never widened. If I want to move it, I close the trade instead.' You deploy it Monday and touch nothing else.

Next Friday you run the same prompt and stop-widening is down from 4 to 1. That is a compounding win, and unlike a vague 'be more disciplined,' you have the number to prove it happened.

The two things that quietly kill the loop

First, keep the prompt and the fields identical week to week. The moment you reword the prompt or change your columns, this week is no longer comparable to last week and you lose the one thing that makes the loop work.

Second, keep the AI in diagnosis and yourself in decisions. It clusters your mistakes and helps you word the rule. It does not pick entries, size positions, or approve orders. You own every risk decision, the AI just makes the review faster and more honest.

Where MyLinedChart fits

This loop lives or dies on a consistent record, and that is the part most traders never get right by hand. MyLinedChart exports your chart work, levels, notes, and trades, as one stable structured file every session, which is exactly the clean, comparable input the Friday prompt needs.

And once you have decided on a rule, MyLinedChart's AI connection at MCP toolkit can draw the chart-side of it for you, marking the level or adding the indicator your new control depends on, with the same you-approve-before-it-lands rule the whole loop already follows.

For a deeper version of the weekly workflow, see Claude Code and ChatGPT Codex for Traders: A Weekly Edge-Upgrade Workflow. To try the loop yourself, start at Pricing.

Do I need to know how to code?

No. You log trades in plain fields and paste them to the AI with one fixed prompt. Claude Code and Codex are just doing the reading and clustering, you make every decision.

Should the AI choose my entries?

No. Keep it in diagnosis and rule-design. It finds the pattern; you decide and enforce the rule.

How often should I change the prompt?

Keep it identical through the loop. Stable prompt plus stable fields is what makes each week comparable to the last.

Can I ship more than one rule a week?

You can, but you lose the ability to tell which change worked. One rule at a time keeps the cause and effect clean.

What is the first success metric?

A measurable drop in the specific mistake you targeted, fewer stop-widens, fewer chases, fewer early exits.