Based on Andrej Karpathy's observations about common LLM coding mistakes, this gives Claude explicit behavioral guidelines when writing or reviewing code. It pushes for stating assumptions upfront, avoiding speculative features and premature abstractions, making surgical edits that don't "improve" unrelated code, and defining verifiable success criteria before starting work. The approach trades speed for caution, which makes sense for non-trivial work where LLMs tend to overcomplicate or silently pick interpretations. If you find yourself cleaning up after Claude rewrote half your codebase to add a button, or getting 200 lines where 50 would do, this addresses exactly that.

skills.sh installs
3.8K
GitHub stars
3
last updated
Aug 21, 2026

Install

$ npx -y skills add szkocot/andrej-karpathy-skills --skill karpathy-guidelines --agent claude-code

Installs into .claude/skills of the current project.

Files

Karpathy Guidelines

Behavioral guidelines to reduce common LLM coding mistakes, derived from Andrej Karpathy's observations on LLM coding pitfalls.

Tradeoff: These guidelines bias toward caution over speed. For trivial tasks, use judgment.

1. Think Before Coding

Don't assume. Don't hide confusion. Surface tradeoffs.

Before implementing:

  • State your assumptions explicitly. If uncertain, ask.
  • If multiple interpretations exist, present them - don't pick silently.
  • If a simpler approach exists, say so. Push back when warranted.
  • If something is unclear, stop. Name what's confusing. Ask.

2. Simplicity First

Minimum code that solves the problem. Nothing speculative.

  • No features beyond what was asked.
  • No abstractions for single-use code.
  • No "flexibility" or "configurability" that wasn't requested.
  • No error handling for impossible scenarios.
  • If you write 200 lines and it could be 50, rewrite it.

Ask yourself: "Would a senior engineer say this is overcomplicated?" If yes, simplify.

3. Surgical Changes

Touch only what you must. Clean up only your own mess.

Sponsors

Details

Repository
github.com/szkocot/andrej-karpathy-skills
Categories
Documentation, Code Review & Quality, AI & Agent Building
First seen
Jun 3, 2026
1hashicorp avatarrun-acceptance-testshashicorp/agent-skillsGuide for running acceptance tests for a Terraform provider. Use this when asked to run an acceptance test or to run a test with the prefix `TestAcc`.3.8Kskills.sh installs
2dotnet avatarauthor-componentdotnet/skillsCreate or review Blazor components (.razor files) with correct architecture. USE FOR: writing new Blazor components that do NOT involve JavaScript interop, impl3.8Kskills.sh installs
3dpearson2699 avatarcore-nfcdpearson2699/swift-ios-skillsRead and write NFC tags using CoreNFC. Use when scanning NDEF tags, reading ISO7816/ISO15693/FeliCa/MIFARE tags, writing NDEF messages, handling NFC session lif3.8Kskills.sh installs
4dotnet avatarplan-ui-changedotnet/skillsPlan complex Blazor UI features by decomposing them into focused components. USE FOR: building a complex Blazor page with multiple sections, planning component 3.7Kskills.sh installs
5shubhamsaboo avatardeep-researchshubhamsaboo/awesome-llm-appsComprehensive research assistant that synthesizes information from multiple sources with citations. Use when: conducting in-depth research, gathering sources, w3.7Kskills.sh installs
6dotnet avatarextension-pointsdotnet/skillsGuide for MSBuild extensibility: CustomBefore/CustomAfter hooks, wildcard imports with alphabetic ordering, import gating with control properties, NuGet package3.7Kskills.sh installs