Have an AI write scripts for you. The whole reference comes as one text file, and every problem the editor reports has a stable code a model can act on.
The reference for an AI
terminal/llms.txt is the whole scripting reference in one file: these guides, every diagnostic code, the API’s TypeScript declarations, every example and every built-in indicator. It is built from the same sources as these pages and the engine, so it matches what the terminal runs. Give it to a model with your request, then paste the script into the editor.
What to ask for
A model writes a better script when the request says what it should read and draw. Name:
- the data: the chart’s bars, another interval or market, the tape, liquidations, liquidation levels, the order book, open interest, funding or breadth;
- what to plot and where: on the price pane, in a pane of its own, as a line, bars, markers or a table;
- the settings the reader should be able to change;
- what should happen on the bar still forming: follow the price live, or wait for the bar to close.
For example: “Using the 0xArchive scripting reference, write a script that plots, in its own pane, the USD of long liquidations less short liquidations on each bar, with a setting for the smallest liquidation to count, and marks bars where more than $5M of longs were liquidated.”
The rules a model most often breaks are in how a script runs: constructors only at the top, persistent values as declared state, and indicators updated on every bar.
Fixing what it wrote
The editor checks a script as you type and lists each problem with its line and a code, such as OX1003 for a top-level let. Give the model the list: each code’s entry in diagnostics says what is wrong and what to do. Codes never change meaning, so a fix-up loop can key on them. OX9xxx codes are the engine’s own failures, never the script’s.
Next step
Look up a code in diagnostics. Last modified on October 16, 2026
