karpathy-principles

Reduce common LLM coding mistakes by enforcing Karpathy-inspired behavioral guidelines like Think Before Coding and Surgical Changes. These principles structure decision-making and apply explicit rules to planning, implementing, and reviewing non-trivial coding tasks.

Disciplined AI coding workflows benefit from Simplicity First, Goal-Driven Execution, and minimal purpose-built changes. Transforming tasks into verifiable steps with explicit success criteria ensures safe, auditable code generation across AI agents.

Apply a safety checklist to non-trivial coding tasks by surfacing assumptions, requiring minimal purpose-built changes, and touching only necessary code. This goal-driven execution transforms tasks into verifiable steps with explicit success criteria.

Enforcing coding discipline prevents unnecessary refactors in AI agents by requiring Surgical Changes. This rule ensures you touch only what you must, avoiding changes to unrelated code and keeping modifications minimal and purpose-built.

Use goal-driven execution for LLM coding tasks when implementing non-trivial code across AI agents. It transforms tasks into verifiable steps with explicit success criteria, reducing unsafe coding practices through structured, auditable decision-making.