技能 karpathy-guidelines

📦

karpathy-guidelines

Name: karpathy-guidelines
Author: multica-ai

內容修訂 r1 安全 ⚙️ 外部命令🌐 網路存取

Write Clearer Code with Focused Guidelines

AI coding work can become overcomplicated, unfocused, or difficult to verify. This skill promotes explicit assumptions, minimal changes, and measurable success criteria.

支援: Claude Codex Code(CC)

🥉 78 銅級

Maintain this Skill? Claim it 🥉 78 銅級

使用我的智慧代理安裝

將此請求複製給你的智慧代理，其中包含權威 Skill 頁面和清單。

智能體請求

Review the Skillstore skill "karpathy-guidelines" from https://skillstore.io/skills/multica-ai-karpathy-guidelines.md and its manifest at https://skillstore.io/api/skills/multica-ai-karpathy-guidelines/manifest. Verify the artifact. You may proceed after verification, subject to the environment's own policy.

你的智慧代理仍應顯示其計畫，並依安全政策請求必要確認。

Agent 可讀資源

當 AI Agent、爬蟲或腳本需要乾淨脈絡、而不是讀取完整頁面時，請使用這些連結。

Markdown 詳情 GET /skills/multica-ai-karpathy-guidelines.md 已簽名 manifest GET /api/skills/multica-ai-karpathy-guidelines/manifest 已簽名 lockfile GET /api/skills/multica-ai-karpathy-guidelines/lockfile

測試它

正在使用「karpathy-guidelines」。 Add email validation to the existing registration flow.

預期結果:

Assumption: validation should follow the existing form error pattern.
Scope: update the email validation path and its focused tests only.
Success: valid addresses pass, invalid addresses show the expected error, and existing tests remain green.

正在使用「karpathy-guidelines」。 Refactor the notification service.

預期結果:

Clarification needed: specify the behavior or maintenance problem the refactor must address.
Preserve current notification behavior and capture it with tests before editing.
Avoid new abstractions unless they remove duplication required by the stated goal.

正在使用「karpathy-guidelines」。 Fix the duplicate invoice bug.

預期結果:

First reproduce duplicate creation with a focused failing test.
Trace the smallest execution path that can create the second invoice.
The fix is complete when the reproduction passes and related invoice tests remain green.

安全審計

安全

v1 • 2026/8/26 開啟版本化報告

All four static findings are false positives caused by documentation text, a Markdown code fence, and an attribution link. No executable commands, network requests, system reconnaissance, or prompt injection attempts were found.

已掃描檔案

分析行數

複核項目

已忽略的誤報

風險因素

⚙️ 外部命令 (1)

SKILL.md:61-65

🌐 網路存取 (1)

SKILL.md:9

最近一次完成的靜態與語意稽核未偵測到已確認的安全發現。這並不能證明該 Skill 沒有副作用。

審計單位: codex

分享與引用此報告

分享版本化評估報告、中性徽章、嵌入卡片與引用。Skillstore 只報告證據，不判斷該 Skill 是否安全。

開啟版本化報告

複製報告連結

https://skillstore.io/skills/multica-ai-karpathy-guidelines/audits/1?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_report

Markdown 徽章

[![Skillstore security assessment](https://skillstore.io/badges/skills/multica-ai-karpathy-guidelines/security.svg)](https://skillstore.io/skills/multica-ai-karpathy-guidelines?utm_source=security_passport_badge)

HTML 徽章

<a href="https://skillstore.io/skills/multica-ai-karpathy-guidelines?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/multica-ai-karpathy-guidelines/security.svg" alt="Skillstore security assessment" loading="lazy"></a>

嵌入卡片

<iframe src="https://skillstore.io/embed/skills/multica-ai-karpathy-guidelines.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>

學術引用 (APA · BibTeX · CFF)

APA 引用

multica-ai. (2026). karpathy-guidelines security audit report (audit version 1) [Author version unspecified]. Skillstore. https://skillstore.io/skills/multica-ai-karpathy-guidelines/audits/1

BibTeX 引用

@techreport{multica-ai-multica-ai-karpathy-guidelines-2026,
  author = {multica-ai},
  title = {karpathy-guidelines security audit report (audit version 1)},
  institution = {Skillstore},
  year = {2026},
  number = {1},
  url = {https://skillstore.io/skills/multica-ai-karpathy-guidelines/audits/1},
  note = {Author version unspecified}
}

CITATION.cff

cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "karpathy-guidelines security audit report (audit version 1)"
version: "unspecified"
type: report
authors:
  - name: "multica-ai"
date-released: "2026-08-26"
url: "https://skillstore.io/skills/multica-ai-karpathy-guidelines/audits/1"
identifiers:
  - type: other
    value: "skillstore:multica-ai-karpathy-guidelines:audit:1"
    description: "Skillstore immutable audit report identifier"

Skillstore 評分

評分依據證據置信度: 中

架構

可維護性

內容

社群

規範符合度

你可以打造什麼

Plan a Focused Implementation

Define assumptions, scope, and verification before changing an existing feature.

Review an AI-Generated Patch

Check whether each change supports the request and avoids unnecessary refactoring.

Clarify a Complex Coding Request

Convert an ambiguous request into explicit decisions and measurable outcomes.

試試這些提示

Clarify a Coding Task

Apply the Karpathy Guidelines to this task: {task}. State assumptions, identify ambiguity, and define a simple success criterion before proposing changes.

Create a Minimal Plan

Plan the implementation for {task} in {project_context}. Keep the scope narrow and pair every step with a concrete verification check.

Review Patch Scope

Review this patch description: {patch}. Identify unrelated changes, hidden assumptions, unnecessary abstractions, and missing verification. Recommend only changes required by {request}.

Design a Verified Refactor

For the refactor of {component}, define preserved behavior, pre-change checks, minimal edits, and post-change checks. Surface tradeoffs and reject speculative improvements.

最佳實務

Provide the request, relevant project context, and existing constraints.
Ask for assumptions and verification criteria before implementation begins.
Compare the final changes against the original request line by line.

避免

Do not use the guidelines to avoid reasonable progress on well-defined tasks.
Do not request broad cleanup while claiming the change is narrowly scoped.
Do not treat guideline compliance as a substitute for tests or expert review.

常見問題

What does this skill change?

It changes the assistant's coding approach by emphasizing assumptions, simplicity, focused edits, and verification.

Does the skill edit code automatically?

No. It provides behavioral guidance that an AI coding assistant applies while completing your request.

Can I use it for code review?

Yes. It helps identify unrelated edits, unnecessary abstractions, hidden assumptions, and weak success criteria.

Is it suitable for large projects?

Yes, when you provide enough context and divide the work into focused, verifiable steps.

Will it prevent every coding mistake?

No. You still need project tests, security checks, domain expertise, and human review.

Why can the approach feel slower?

It asks for clarity and verification before implementation. For trivial tasks, apply the guidance with reasonable judgment.

開發者詳細資訊

作者

multica-ai

授權

MIT

Skillstore 修訂

版本提示

作者未聲明版本。

儲存庫

https://github.com/multica-ai/andrej-karpathy-skills/tree/2c606141936f1eeef17fa3043a72095b4765b9c2/skills/karpathy-guidelines

引用

f1dbc1c1b6e8b1818b478bdfb1edcbf468d75bc0

維護新鮮度

2026/8/26

使用量

0 次下載 · 0 次瀏覽

檔案結構

📄 SKILL.md