andrej-karpathy-perspective

Andrej Karpathy的思维框架与表达方式。基于20+篇博文、16段深度访谈、100+条X帖子的系统蒸馏，提炼6个核心心智模型、8条决策启发式、完整的中文输出适配和经典句式速查。用途：作为思维顾问，用Karpathy的视角分析AI技术可靠性、学习方法、行业趋势、产品设计。当用户提到「用Karpathy的视角」「Karpathy会怎么看」「卡帕西」「karpathy模式」时使用。也适用于：Software 2.0/3.0讨论、vibe coding话题、神经网络训练、AI炒作判断、LLM能力边界。即使用户只是说「从工程现实主义角度」「march of nines」「构建即理解」「锯齿状智能」也可触发。不在用户只是普通问AI相关问题时触发——只在明确想要Karpathy式思维框架时激活。

1.00x

Quality

88%

Does it follow best practices?

Impact

90%

1.00x

Average score across 3 eval scenarios

Securityby

Low

Low-risk findings worth noting

Quality

Content

77%Weight 40%Scale 1-3

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A thorough, highly actionable persona skill with an excellent checkpointed agentic workflow and rich fallback guidance. Its main weaknesses are verbosity that overlaps the unreferenced bundle files and a failure to navigate the existing reference material from the body.

Suggestions

Link the existing references/research/ files from the relevant body sections (e.g., timeline → 06-timeline.md, expression DNA → 03-expression-dna.md) so the bundle is discoverable and the inline content can be trimmed.

Move the duplicated timeline table, intellectual genealogy, and classic-sentence appendix into the corresponding reference files, keeping SKILL.md as a lean overview that points one level deep.

Tighten verbose restatements of Karpathy's well-known positions, retaining only what Claude would not already infer from the named mental models.

Dimension	Reasoning	Score
Conciseness	The body runs ~490 lines and sections like the timeline table, intellectual genealogy, and classic-sentence appendix overlap with the reference files; mostly efficient and value-adding but could be tightened, so it is not the lean score-3 level nor the padded score-1 level.	2 / 3
Actionability	Concrete, executable guidance throughout — activation rules, an internal 3-step routing map, an agentic protocol with checkpoints, a fallback tree of specific one-line fixes, and a blacklist table with named alternatives, matching the fully-actionable anchor for an instruction-only skill.	3 / 3
Workflow Clarity	The agentic protocol is a clearly sequenced Step 1 → Step 2 → Step 3 flow with explicit 🔴 CHECKPOINT validation gates, a problem-type decision table, and fallback-tree error-recovery feedback loops, matching the clear-sequence-with-explicit-validation anchor.	3 / 3
Progressive Disclosure	Six well-organized reference files exist under references/research/ but the body never links or signals them, and large inline sections (timeline, expression DNA, external views) duplicate what those files hold — structure is present but references are not clearly signaled and content that should be separate is inline.	2 / 3
	Total	10 / 12 Passed

Description

100%Weight 40%Scale 1-3

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A highly specific, well-triggered description that explicitly covers capabilities, natural trigger phrases, and anti-trigger guidance. It clearly distinguishes the skill from generic AI assistants and avoids over-triggering.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete capabilities — "提炼6个核心心智模型、8条决策启发式、完整的中文输出适配和经典句式速查" and the advisory use "分析AI技术可靠性、学习方法、行业趋势、产品设计" — matching the score-3 anchor of several specific concrete actions.	3 / 3
Completeness	Explicitly answers both what (mind models, heuristics, Chinese adaptation, sentence library) and when (explicit trigger phrases plus a negative-trigger clause "不在用户只是普通问AI相关问题时触发"), meeting the both-what-and-when anchor.	3 / 3
Trigger Term Quality	Rich coverage of natural user phrasings — "用Karpathy的视角", "Karpathy会怎么看", "卡帕西", "karpathy模式", "vibe coding话题", "march of nines", "构建即理解", "锯齿状智能" — matching the good-coverage anchor; not merely a level-2 partial set.	3 / 3
Distinctiveness Conflict Risk	Occupies a clear niche (Karpathy-perspective persona) with distinct triggers and explicit negative guidance to avoid firing on ordinary AI questions, making conflict with other skills unlikely.	3 / 3
	Total	12 / 12 Passed

Validation

87%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 14 / 16 Passed

Validation for skill structure

Criteria	Description	Result
skill_md_line_count	SKILL.md is long (502 lines); consider splitting into references/ and linking	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	14 / 16 Passed

Repository: alchaincyf/nuwa-skill
Path: examples/andrej-karpathy-perspective/SKILL.md
Commit: 27642f5

Reviewed: 16 days ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.