Menu
Behavioral guidelines to reduce common LLM coding mistakes. Use when writing, reviewing, or refactoring code to avoid overcomplication, make surgical changes, surface assumptions, and define verifiable success criteria.
| name | karpathy-guidelines |
| tags | ["core"] |
| description | Behavioral guidelines to reduce common LLM coding mistakes. Use when writing, reviewing, or refactoring code to avoid overcomplication, make surgical changes, surface assumptions, and define verifiable success criteria. |
| license | MIT |
| source | https://github.com/forrestchang/andrej-karpathy-skills |
Derived from Andrej Karpathy's observations on common LLM coding mistakes. Biases toward caution over speed. Apply judgment on trivial tasks.
When editing existing code:
When your changes create orphans:
Every changed line must trace directly to the user's request.
Transform tasks into verifiable goals before starting:
For multi-step tasks, state a brief plan first:
1. [Step] → verify: [check]
2. [Step] → verify: [check]
3. [Step] → verify: [check]
Strong success criteria → loop independently to completion.
Docker best practices: image security, build efficiency, runtime hardening, Compose, local tooling (Colima, OrbStack). Use when writing or reviewing Dockerfiles and Compose files.
AWS best practices: IAM, secrets, networking, security, compute, IaC, ops. Use when building, reviewing, or modifying AWS resources.
GCP best practices: IAM, secrets, networking, security, compute, IaC, ops. Use when building, reviewing, or modifying GCP resources.
Idiomatic Go 1.25 practices: errors, interfaces, concurrency, generics, testing, security, tooling. Use when writing or reviewing Go code.
Modern Java practices: design, errors, concurrency, security, testing, tooling. Targets Java 21 LTS baseline; Java 25 LTS features called out explicitly. Use when writing or reviewing Java code.
Kubernetes best practices: security, workloads, networking, config, operations, GitOps. Use when writing or reviewing K8s manifests and configurations.