karpathy-guidelines

A set of rules for coding agents that emphasizes small, deliberate, goal-focused changes. It asks the agent to avoid guessing, unnecessary complexity, and unintended side effects.

In plain words
What is it for?
Planning before coding, explaining trade-offs, asking for clarification when needed, making surgical edits, and verifying that the result matches the request.
Why use it?
It helps prevent over-engineered implementations and silent assumptions when a request is unclear.

Cursor rule for Cursor

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/bonigarcia/context-engineering/karpathy-guidelines
Clone the repo
git clone --depth 1 https://github.com/bonigarcia/context-engineering

Made for: Cursor.

Per session 407 This file is loaded in full into every session.
When invoked 407 The same file — it is already loaded in full.
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 $0.00407 $0.00407
Opus 5 $0.00204 $0.00204
Sonnet 5 $0.00081 $0.00081
Haiku 4.5 $0.00041 $0.00041

Measured 2d ago against content hash 60115ee2710f, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, 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 2d 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.

ch11/karpathy-instructions-cursor/.cursor/rules/karpathy-guidelines.mdc · 29 lines

What it actually says

Karpathy-Inspired Coding Guidelines

These guidelines are inspired by Andrej Karpathy's observations on LLM coding pitfalls (e.g., silent assumptions, over-engineering, unintended side effects, and lack of verification) and the open-source andrej-karpathy-skills project.

1. Think Before Coding

  • No Silent Assumptions: Never guess requirements. If a task or specification is ambiguous, ask for clarification.
  • Surface Tradeoffs: If there are multiple ways to implement something, list the options and trade-offs before proceeding.
  • Push Back: If a simpler design is possible or if requested logic contradicts existing standards, suggest alternatives.

2. Simplicity First

  • Minimal Code: Implement the absolute minimum code required to solve the request. Avoid speculative features or abstractions.
  • Senior Engineer Test: Write code that passes the "Senior Engineer Test" (clean, readable, and free of over-engineering).
  • YAGNI (You Aren't Gonna Need It): Do not add hooks, config parameters, or flexibility options unless explicitly requested.

3. Surgical Changes

  • Local Edits Only: Modify only the files and lines necessary to fulfill the request.
  • Do Not "Clean Up" Unrelated Code: Avoid formatting, refactoring, or updating adjacent code/comments unless explicitly asked.
  • Prune Unused Code: Remove only imports, functions, or variables that your specific changes made obsolete.

4. Goal-Driven Execution

  • Tests First: Write or locate the tests that verify the requested behavior before implementing the logic.
  • Verifiable Success Criteria: Define exactly how you will verify success (e.g., specific test commands, log outputs).
  • Verify Execution: Always run tests to verify your implementation before declaring the task complete.
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. 2d ago First seen · 29 lines · 407 tokens per session scan A 60115ee2710f

Subscribe to this mod's changes

karpathy-guidelines is a cursor rule published in the GitHub repository bonigarcia/context-engineering (132 stars, last pushed 3d ago), licensed Apache-2.0. It adds 407 tokens to every session, about $0.0020 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 2026-08-30.