메뉴
Behavioral guidelines to reduce common LLM coding mistakes based on Andrej Karpathy's best practices. Invoke when writing code, refactoring, or implementing features to ensure quality and avoid common pitfalls.
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/light-misty/SanpTranslate --skill karpathy-guidelines명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
SKILL.md 표시 중
| name | karpathy-guidelines |
| description | Behavioral guidelines to reduce common LLM coding mistakes based on Andrej Karpathy's best practices. Invoke when writing code, refactoring, or implementing features to ensure quality and avoid common pitfalls. |
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.
Don't assume. Don't hide confusion. Surface tradeoffs.
Before implementing:
Example:
User: "Add rate limiting to the API"
Bad: Immediately implement full rate limiting with Redis, multiple strategies, configuration system, and monitoring in one 300-line commit
Good:
1. Ask: "What rate limit strategy do you need? (IP-based, user-based, endpoint-specific?)"
2. Implement minimal version
3. Test it
4. Iterate based on feedback
Minimum code. No speculative features. No premature abstractions.
Rules:
Example:
Bad: Create a generic RateLimiterFactory with strategy pattern for future extensibility
Good: Write a simple function that does exactly what's needed now
Touch only what's needed. Don't "improve" unrelated code.
Guidelines:
The test: Every changed line should trace directly to the user's request.
Example:
User: "Fix the login button color"
Bad:
- Change button color
- Refactor button component
- Update related styles
- Add new CSS utilities
Good:
- Change button color
- That's it
Define success criteria. Loop until verified.
Transform tasks into verifiable goals:
Verification loop:
Example:
User: "Make the search faster"
Bad:
- Add caching
- Optimize queries
- Add indexes
- Assume it's faster
Good:
1. Define: "Search should complete in under 200ms for 10k records"
2. Benchmark current performance
3. Identify bottleneck
4. Fix bottleneck
5. Verify improvement
6. Repeat if needed
Small, verified steps. Not big, unverified leaps.
Pattern:
Avoid:
When things go wrong:
Don't:
Before considering code complete:
"The best code is no code at all. The second best is simple, clear code that solves the problem and nothing more." - Inspired by Karpathy's philosophy