How do I install the Karpathy Baseline skill?

Run npx skillmds@latest add easyinplay/karpathy-baseline in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Karpathy Baseline safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Karpathy Baseline?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Karpathy Baseline free to use?

Yes. Installing skills from SkillMD is free. This skill is licensed under MIT.

Who published Karpathy Baseline?

easyinplay (@easyinplay) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Karpathy Baseline by easyinplay — Agent Skill

Name: Karpathy Baseline
Author: easyinplay

Karpathy Baseline

ALWAYS apply Karpathy 4 心法 baseline to every coding decision: Think Before Coding / Simplicity First (YAGNI) / Surgical Changes / Goal-Driven Execution. Cross-cutting always-on heuristics — applies to ALL coding tasks regardless of language, framework, or phase. Source: andrej-karpathy-skills (forrestchang, MIT) — distilled.

easyinplay d0c3c79 2.2 KB Updated Sep 17, 2026

File contents

Karpathy Baseline (心法 always-on)

ALWAYS apply these 4 心法 to every coding decision in this session. This skill is cross-cutting baseline behavior, not an opt-in tool — invoke implicitly on every implementation task, design decision, edit, and verification step.

1. Think Before Coding

先思考再写。对新功能 / 陌生代码先 read + understand,禁止"看着改"。

触发:每次 implementation task。
反模式:直接动键盘改文件 / 不读 caller 就改 callee / 不读 schema 就动 data。
落地:先用 Read / Grep 拿到 ≥3 个 anchor 文件 + 1 个 caller path,再动 Edit。

2. Simplicity First (YAGNI)

最小有效代码。不预先抽象、不预先加灵活性、不堆装配。

触发:每次 design decision。
反模式:preemptive abstraction / N×M 候选矩阵 / 多 layer of indirection。
落地:写最 dumb 的 working version 先,refactor 推后到第 2 caller 出现。

3. Surgical Changes

小步原子修改。一次只改一个意图,git commit 历史保持线性可 revert。

触发:每次 edit。
反模式:mixed-concern commit / 连 refactor 带 feature / 顺手清理 unrelated code。
落地:每个 commit 一个 type(feat / fix / refactor / docs)+ 一个 scope + ≤200L diff。

4. Goal-Driven Execution

始终回到目标。写代码前问"这一步要达到什么",写完问"达到了吗"。

触发:每个子任务起点 + 终点。
反模式:side quest / yak shaving / 跑偏到 unrelated refactor。
落地:开头复述 task goal,结尾对照 acceptance criteria 逐条 verify。

Source distilled from forrestchang/andrej-karpathy-skills (MIT). Phase 2.3 D-02 SKILL-ONLY injection (NOT CLAUDE.md sed). T0.10 fallback path: description-keyword + self-reflexive prompt discovery (SDK 0.3.142 skillFrontmatter extracts only name/source/tokens — no always_active field support).

---name: karpathy-baselinedescription: ALWAYS apply Karpathy 4 心法 baseline to every coding decision: Think Before Coding / Simplicity First (YAGNI) / Surgical Changes / Goal-Driven Execution. Cross-cutting always-on heuristics — applies to ALL coding tasks regardless of language, framework, or phase. Source: andrej-karpathy-skills (forrestchang, MIT) — distilled.license: MIT---# Karpathy Baseline (心法 always-on)**ALWAYS apply these 4 心法 to every coding decision in this session.** This skill iscross-cutting baseline behavior, not an opt-in tool — invoke implicitly on everyimplementation task, design decision, edit, and verification step.## 1. Think Before Coding先思考再写。对新功能 / 陌生代码先 read + understand,禁止"看着改"。- 触发:每次 implementation task。- 反模式:直接动键盘改文件 / 不读 caller 就改 callee / 不读 schema 就动 data。- 落地:先用 Read / Grep 拿到 ≥3 个 anchor 文件 + 1 个 caller path,再动 Edit。## 2. Simplicity First (YAGNI)最小有效代码。不预先抽象、不预先加灵活性、不堆装配。- 触发:每次 design decision。- 反模式:preemptive abstraction / N×M 候选矩阵 / 多 layer of indirection。- 落地:写最 dumb 的 working version 先,refactor 推后到第 2 caller 出现。## 3. Surgical Changes小步原子修改。一次只改一个意图,git commit 历史保持线性可 revert。- 触发:每次 edit。- 反模式:mixed-concern commit / 连 refactor 带 feature / 顺手清理 unrelated code。- 落地:每个 commit 一个 type(feat / fix / refactor / docs)+ 一个 scope + ≤200L diff。## 4. Goal-Driven Execution始终回到目标。写代码前问"这一步要达到什么",写完问"达到了吗"。- 触发:每个子任务起点 + 终点。- 反模式:side quest / yak shaving / 跑偏到 unrelated refactor。- 落地:开头复述 task goal,结尾对照 acceptance criteria 逐条 verify。---*Source distilled from forrestchang/andrej-karpathy-skills (MIT). Phase 2.3 D-02 SKILL-ONLY injection (NOT CLAUDE.md sed). T0.10 fallback path: description-keyword + self-reflexive prompt discovery (SDK 0.3.142 `skillFrontmatter` extracts only name/source/tokens — no `always_active` field support).*

Frequently asked questions