[QUESTION]: If Karpathy's LLM coding principles are "obviously right," why isn't Claude Code already shipping them by default?

[xpolb01]

There's a repo floating around with ~30k stars — andrej-karpathy-skills — that packages four principles from Karpathy's tweet into a single CLAUDE.md:

1. Think Before Coding — surface assumptions, don't silently pick
2. Simplicity First — minimum code, no speculative abstraction
3. Surgical Changes — don't drive-by refactor
4. Goal-Driven Execution — transform tasks into verifiable goals

Here's the thing I can't stop thinking about: most of these are already in Claude Code's default system prompt. If you inspect what CC actually sends, there's language like "Don't add features beyond what the task requires" and "A bug fix doesn't need surrounding cleanup" and "Don't add error handling for scenarios that can't happen."

So the defaults already lean this way. And yet this repo exists, has 30k stars, and people feel the need to install it on top.

Why? A few theories I'm torn between:

• The defaults are soft nudges and get drowned out the moment your own prompt contradicts them
• "Goal-Driven Execution" (the TDD-reframe trick) is the one principle that's genuinely missing from defaults
• Anthropic tunes for "feels helpful to the median user," which mildly conflicts with surgical restraint
• Every model release quietly absorbs more of this, and the repo will age out in a year

Curious what others are seeing. For people using the Karpathy prompt: which principle actually changed Claude's behavior, and which was already being honored? And has anyone seen Anthropic explain the design trade-off?

Not a criticism — genuine puzzle.