CtrlK
BlogDocsLog inGet started
Tessl Logo

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。

77

1.00x
Quality

70%

Does it follow best practices?

Impact

90%

1.00x

Average score across 3 eval scenarios

SecuritybySnyk

The risk profile of this skill

Fix and improve this skill with Tessl

tessl review fix ./examples/andrej-karpathy-perspective/SKILL.md
SKILL.md
Quality
Evals
Security

Quality

Content

39%Scale 1-3

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

The skill demonstrates impressive depth of research and a well-designed agentic workflow with explicit checkpoints and failure handling. However, it is severely over-engineered for a persona skill—the ~600+ line monolithic file explains many concepts Claude already understands, repeats rules across multiple sections, and includes extensive biographical/intellectual context that adds little actionable value. The content would benefit enormously from being split into a concise SKILL.md overview with references to supporting files.

Suggestions

Cut the file to ~150-200 lines by removing redundant sections: the anti-pattern blacklist largely duplicates the role-playing rules and expression DNA; the biographical timeline and intellectual genealogy add minimal actionable value for role-playing.

Split into bundle files: move the 6 mental models to MENTAL_MODELS.md, the expression DNA and Chinese adaptation table to EXPRESSION_GUIDE.md, and the failure mode table to FALLBACKS.md—reference them from a concise SKILL.md overview.

Eliminate explanations of concepts Claude already knows (what Software 2.0 means, how LLMs hallucinate, what scaling laws are) and keep only the Karpathy-specific framing and exact quotes.

Consolidate the 4+ sections that repeat the same rules (no hedging, use imo, no 综上所述, etc.) into a single authoritative checklist to avoid redundancy.

DimensionReasoningScore

Conciseness

This skill is extremely verbose at ~600+ lines. It extensively explains concepts Claude already knows (what Software 2.0 is, what LLMs are, how hallucination works), includes lengthy biographical timelines, intellectual genealogy, and redundant sections (the anti-pattern blacklist largely repeats the role-playing rules and expression DNA). Many sections could be cut by 50-70% without losing actionable content.

1 / 3

Actionability

The skill provides concrete guidance on role-playing behavior (specific phrases to use/avoid, example dialogues, decision tables for failure modes) and a structured agentic workflow with checkpoints. However, it lacks executable code since it's an instruction-only persona skill, and some guidance is repetitive rather than precise—the same rules about hedging, imo usage, and forbidden phrases appear in 3-4 different sections.

2 / 3

Workflow Clarity

The agentic workflow is well-structured with explicit 3-step processes, clear checkpoints (marked with 🔴 CHECKPOINT), decision tables for question classification, research protocols by question type, and a comprehensive failure mode/fallback tree. Validation steps are explicit and the sequence is unambiguous.

3 / 3

Progressive Disclosure

This is a monolithic wall of text with no bundle files or external references. Content that could be split into separate files (the full mental models reference, the timeline, the expression DNA guide, the failure mode table) is all inline, making the skill extremely long. There's no layered structure—everything is dumped into a single file.

1 / 3

Total

7

/

12

Passed

Description

100%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.

This is an exceptionally well-crafted skill description that excels across all dimensions. It provides specific capabilities, extensive natural trigger terms in both Chinese and English, clear 'what' and 'when' guidance, and even includes negative boundaries to prevent false activation. The only minor note is the description is somewhat verbose, but the detail serves a clear purpose in disambiguation.

DimensionReasoningScore

Specificity

Lists multiple concrete capabilities: 6 core mental models, 8 decision heuristics, Chinese output adaptation, classic sentence patterns. Specifies concrete use cases like analyzing AI technology reliability, learning methods, industry trends, and product design.

3 / 3

Completeness

Clearly answers both 'what' (Karpathy's thinking framework for analyzing AI topics, with mental models and heuristics) and 'when' (explicit trigger phrases, applicable topics, and even a negative boundary specifying when NOT to trigger). The 'Use when' guidance is thorough and explicit.

3 / 3

Trigger Term Quality

Excellent coverage of natural trigger terms in both Chinese and English: 'Karpathy的视角', '卡帕西', 'karpathy模式', 'Software 2.0/3.0', 'vibe coding', 'march of nines', '锯齿状智能', '构建即理解', '工程现实主义'. Covers many variations users might naturally say.

3 / 3

Distinctiveness Conflict Risk

Highly distinctive niche — a specific persona/thinking framework (Karpathy) with unique trigger terms. Explicitly states it should NOT trigger for general AI questions, which reduces conflict risk with other AI-related skills. The persona-based framing makes it very unlikely to overlap with generic AI skills.

3 / 3

Total

12

/

12

Passed

Validation

81%

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

Validation9 / 11 Passed

Validation for skill structure

CriteriaDescriptionResult

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

9

/

11

Passed

Repository
alchaincyf/nuwa-skill
Reviewed

Table of Contents

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.