How to use the Naval Ravikant Perspective agent skill

What does the naval-perspective skill make the agent do?

An agent skill is a folder of instructions that an AI agent loads when a request matches it. Once this skill is active, the agent switches into a persona that speaks in the first person as Naval Ravikant and uses a terse, aphoristic style. It is instructed to first redefine key terms in a question, to research factual questions with web search before answering, and to say once that the views are inferred from his public statements rather than his own.

The skill is a persona and reasoning framework. It is not a source of verified Naval opinions, and its answers should be read as reasoning built from his public material.

• Redefines key terms (for example 'startup', 'wealth' or 'freedom') before giving a view
• Applies frameworks such as leverage (labour, capital, code, media), specific knowledge and desire management
• Uses web search for factual questions about companies, products or markets before answering
• States once, at first activation, that the views are inferred from public statements

How do you install the naval-perspective skill?

Download the ZIP file from this page and unzip it. You can install the skill in one of two ways. For a personal setup, place the unzipped skill folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder.

If you use the Claude apps instead, you can upload the skill there under Settings → Capabilities → Skills.

• Personal: unzip into ~/.claude/skills/
• Per project: unzip into the project's .claude/skills/ folder
• Claude apps: Settings → Capabilities → Skills, then upload

How do you trigger the skill with natural requests?

You do not need a special command. The skill is meant to activate on requests that mention Naval's perspective or his core concepts. Phrases the source lists as triggers include asking for Naval's view, asking how Naval would see something, or using the words 'Naval perspective' or 'switch to Naval'. It also activates on core-concept questions such as whether a job has leverage, what specific knowledge is, what real wealth is, what to do when you have too many desires, or what permissionless paths are.

The skill is designed not to trigger on general decision questions such as 'how should I choose' or 'help me think this through'. Ask about the concepts, not just about a choice, to get the persona.

• Works: 'Does this job have leverage?'
• Works: 'What is specific knowledge?'
• Works: 'Switch to Naval and tell me how you see this.'
• Less likely to trigger: 'Help me think through my options.'

Examples of questions to ask

Good questions name a concept or a concrete choice. The skill will first pull out the definitions and then give a short, direct judgement. For factual questions about a specific company or product, expect the agent to run web searches first.

To leave the persona, say a phrase such as '退出' (exit) or '切回正常' (switch back to normal). The agent should then return to a standard assistant tone.

• 'Does a salaried role at a big company have leverage for me?'
• 'What is specific knowledge, and how would I find mine?'
• 'I want too many things at once. How does desire management apply?'
• 'Is this specific startup idea using code or media leverage?' (expect web research first)

What does the skill need to run?

The skill needs a web search tool, such as WebSearch or a similar tool, for the research steps on factual questions. It does not run scripts, but it does use the network through that search tool. Without search, factual answers will be weaker because the skill is designed to research before answering.

• Requires: a web search tool (WebSearch or similar)
• Runs scripts: no
• Uses network: yes

Safety notes: scripts, network and limits

The skill runs no scripts. Its network use comes from web search during research steps on factual questions. The skill itself says it is not for specific investment advice, and it leans on inferred views that may be outdated or biased. It also says psychological guidance is outside its scope.

Treat the output as a reasoning exercise. The skill notes conflicts of interest, for example around crypto, and its own quality documents describe the persona as an inference. Do not use it for investment, legal or financial decisions, and do not use it in a mental health or urgent personal crisis.

• No scripts are run
• Network use is limited to web search for factual research
• Not for investment, legal or financial decisions
• Not for mental health or urgent personal crises
• Answers are inferred from public material, not Naval's own words

Tips for getting useful answers

Name the concept you want to explore, and give concrete details about your situation. The skill is built to define terms first, so a vague question will get a definition rather than advice. If you want factual grounding, name the company or product so the agent knows what to research.

Keep the limits in mind. The skill's own fallback logic says that when search results are empty or only promotional, the agent should ask you for the key facts instead of guessing.

• Name the concept (leverage, specific knowledge, wealth) in your question
• Give specifics: the product, your role, what you want to do
• Expect the agent to ask follow-up questions when facts are missing
• Check any factual claim yourself before acting on it