karpathy-guidelines

karpathy-guidelines is a cursor rule for Cursor from dasomel/oh-my-cursor. It costs 19 tokens per session (484 once invoked), scanned A, original, MIT.

A set of coding rules focused on understanding the problem, keeping solutions simple, making small targeted changes, and verifying them with tests.

In plain words
What is it for?
Use it to guide bug fixes, feature changes, code reviews, and decisions about how much code to write or change.
Why use it?
It reduces unnecessary code, accidental refactoring, and assumptions that can lead to bugs.

Cursor rule for Cursor

Written for Cursor: a Cursor rule (.mdc).

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add rules/dasomel/oh-my-cursor/karpathy-guidelines
Clone the repo
git clone --depth 1 https://github.com/dasomel/oh-my-cursor

Made for: Cursor.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for karpathy-guidelines

README.md
[![agentmods](https://agentmods.dev/badge/rules/dasomel/oh-my-cursor/karpathy-guidelines.svg)](https://agentmods.dev/rules/dasomel/oh-my-cursor/karpathy-guidelines)
Your own site
<a href="https://agentmods.dev/rules/dasomel/oh-my-cursor/karpathy-guidelines"><img src="https://agentmods.dev/badge/rules/dasomel/oh-my-cursor/karpathy-guidelines.svg" alt="Measured on agentmods" height="20"></a>
Per session 19 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 484 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00019 $0.00484
Opus 5 $0.00010 $0.00242
Sonnet 5 $0.00004 $0.00097
Haiku 4.5 $0.00002 $0.00048

Measured 5d ago against content hash ad0b039d4414, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-06, from the pricing page.

Security

Grade A, and why

karpathy-guidelines scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

rules/practices/karpathy-guidelines.mdc · 62 lines

What it actually says

Karpathy Guidelines

Coding discipline inspired by Andrej Karpathy's principles.

Core Principles

1. Think Before Coding

  • Understand the problem completely before writing any code
  • If you can't explain the approach in plain language, you're not ready to code
  • 10 minutes of thinking saves 2 hours of debugging

2. Simplicity Above All

  • The best code is code you didn't write
  • Every line of code is a liability — minimize it
  • If a junior developer can't understand it, it's too complex
  • Choose boring technology over clever solutions

3. Surgical Changes

  • Change the minimum needed to solve the problem
  • Don't "clean up" code around your change
  • Don't refactor while fixing bugs
  • One PR, one purpose

4. Verify Everything

  • Don't trust your assumptions — run the code
  • Write the test that would have caught this bug
  • If you can't test it, you can't ship it
  • "It works on my machine" is not verification

5. Read More Code Than You Write

  • Read the existing code before changing it
  • Understand why the code is the way it is
  • The previous developer might have had a good reason
  • Chesterton's Fence: don't remove something until you understand why it's there

Anti-Patterns to Avoid

  • Resume-Driven Development — using fancy tech to learn it, not because it's needed
  • Premature Abstraction — "we might need this later" (you won't)
  • Config-Driven Everything — hardcode first, extract config only when needed
  • Framework Tourism — switching frameworks instead of solving the actual problem

Decision Framework

Before writing code, ask:

  1. Can I solve this without code? (config change, removal, documentation)
  2. Can I solve this with less code? (simpler algorithm, existing library)
  3. Will someone else understand this in 6 months?
  4. Am I solving the actual problem or a hypothetical one?

Rules

  • Delete code whenever possible — less is more
  • Prefer explicit over implicit
  • Prefer flat over nested
  • Prefer simple over clever
Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 5d ago First seen · 62 lines · 19 tokens per session scan A ad0b039d4414

Subscribe to this mod's changes

karpathy-guidelines is a cursor rule published in the GitHub repository dasomel/oh-my-cursor (2 stars, last pushed 6mo ago), licensed MIT. It adds 19 tokens to every session and 484 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.