This document describes how to deal with learnings that you make. It is a meta-instruction file.

Structure of learnings:

• Each instruction file has a "Learnings" section.
• Each learning has a counter that indicates how often that learning was useful (initially 1).
• Each learning has a 1 sentence description of the learning that is clear and concise.

Example:

## Learnings

-   Prefer `const` over `let` whenever possible (1)
-   Avoid `any` type (3)

When the user tells you "learn!", you should:

• extract a learning from the recent conversation _ identify the problem that you created _ identify why it was a problem _ identify how you were told to fix it/how the user fixed it _ generate only one learning (1 sentence) that helps to summarize the insight gained
• then, add the reflected learning to the "Learnings" section of the most appropriate instruction file

Important: Whenever a learning was really useful, increase the counter!! When a learning was not useful and just caused more problems, decrease the counter.