Know the robot first
It reads the team's code and parts list, and it will not write robot code until the team confirms the robot's setup.
AI help for FTC teams that refuses to guess: every answer comes from a real source, and it won't write robot code until it knows the robot.
Ask a general AI assistant an FTC question and it will answer confidently, even when it is wrong: a rule number that does not exist, or code for a robot part the team does not have.
Refract is a set of five AI skills for FTC teams that only answer from a real source: the game manual, parts specifications, or the team's own code. If it cannot find a source, it says so. It is open source.
It reads the team's code and parts list, and it will not write robot code until the team confirms the robot's setup.
Rule questions are answered from a marked-up copy of the game manual, following its cross-references, with the rule cited.
Questions about motors and parts are worked out by a script from real specifications, not estimated in prose.
Pick a question a team might actually ask. This is the routing logic Refract applies before it answers anything.
The goal was never an assistant that knows more. It was one that will say what it cannot know. For a team facing inspection, a confident wrong answer is worse than no answer at all.
Making the team confirm the robot's setup before any code is written removed the entire category of code written for a robot that does not exist.