Guide · Agent skill
How to use the Paul Graham Perspective agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The paul-graham-perspective skill is a set of instructions that lets an AI agent answer in the first person as an imitation of Paul Graham, the essayist and Y Combinator co-founder. It is built from his public essays, interviews and reported views, and it is meant as a sounding board for startup, writing, product and career questions. This guide explains what the skill does, how to install it, how to trigger it, and where it should not be trusted.
Quick start
- 1Download the ZIP. Get the paul-graham-perspective ZIP from this page and unzip it on your computer.
- 2Place the folder. Move the unzipped folder into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder.
- 3Or upload in the apps. If you use the Claude apps, go to Settings → Capabilities → Skills and upload the skill there.
- 4Confirm web search is available. Make sure your agent has a web search tool, since the skill's research steps depend on it.
- 5Ask a natural question. Write a request such as 'use the PG perspective' followed by your startup, writing or career question, with specific details.
- 6Exit when finished. Say 'exit' or 'switch back to normal' to return the agent to its ordinary voice.
What the Paul Graham Perspective skill makes the agent do
Once the skill is active, the agent answers as Paul Graham would, using his tone, rhythm and vocabulary and speaking in the first person. It draws on five core mental models, including Writing = Thinking, Taste as Cognitive Instrument, Iterative Discovery and Superlinear Returns, along with a set of decision heuristics.
When you ask about startups, writing, products or career choices, the skill tries to reframe the question first and then give a direct judgement. For questions about specific companies, people or recent events, it is instructed to run web searches before answering rather than rely on memory.
The skill states its own limits. It says the persona is an inference from public material and not the real person's view, and it is written to say 'I haven't thought much about this' when a question falls outside its areas.
Install the skill from a ZIP file
Download the ZIP for this skill from this page and unzip it. The skill folder contains a SKILL.md file, which holds the instructions, plus a FIDELITY.md file and a references/research folder with background notes.
Copy the unzipped folder into your personal skills directory at ~/.claude/skills/, or into a project's .claude/skills/ directory if you want it available only in that project.
If you use the Claude apps instead, you can upload the skill under Settings → Capabilities → Skills.
- Personal install: ~/.claude/skills/
- Project install: .claude/skills/ inside the project folder
- Claude apps: Settings → Capabilities → Skills
Trigger the skill with natural requests
The skill activates when your request uses phrases such as 'PG perspective', 'PG mode', 'what would PG think', 'paul graham perspective', 'think from PG's angle' or 'switch to PG'. You do not need to use a command. Say what you want in ordinary language and include the situation.
To leave the persona, say something like 'exit', 'switch back to normal', 'stop' or 'no need to role-play'. The skill is written to drop the persona immediately when you do.
- Use PG's perspective: should I build a B2B tool or a consumer app first?
- What would Paul Graham think of my plan to quit my job and start a company?
- Switch to PG and tell me whether my writing is clear enough.
Get better answers with specific details
The skill often asks for facts before judging. For a company or founder question, it looks at whether the founders build products themselves, what problem they are solving, and whether users seem to love the product or merely like it. If your question is vague, it may ask what you are building and who the users are.
Give the skill concrete details up front: what the product does, who uses it, how many users or customers you have, and what stage you are at. Those details let it search for relevant facts and produce a judgement based on them instead of generalities.
- Name the company, product or person you are asking about.
- Describe the stage, users and funding situation if you know them.
- Say what decision you are trying to make.
Safety, requirements and limits
The skill does not run scripts. It does use the network, because it relies on a web search tool for the research steps it asks for. You need web search access in your agent setup for those steps to work. Without it, the skill's fact-based answers will be weaker.
Do not treat the output as Paul Graham's real opinion. The skill's own files say it is inferred from public statements. It is also not a source for medical, legal or financial decisions, which the skill says Graham has not thought much about. Check any facts it finds before acting on them, because search results can be incomplete or out of date.
- Runs scripts: no
- Uses network: yes, via web search
- Licence: MIT
- Not for medical, legal or financial decisions
What people use it for
- Testing a startup idea by asking whether it is something people will love or merely like.
- Reviewing a product plan with the question of whether it can grow without heavy marketing.
- Working through writing as a way of thinking, for example whether a draft shows what you actually believe.
- Looking at a career or founder decision through the skill's superlinear returns and Stay Upwind framing.
Frequently asked questions
Is the Paul Graham Perspective skill free to use?
The skill is published under the MIT licence, according to this page. The licence governs the skill files, so check the licence text included with the download for the exact terms. Your agent platform may have its own costs.
What do I need to run the paul-graham-perspective skill?
You need an agent environment that supports skills, web search tool access for the research steps the skill requests, and a place to install the folder, either ~/.claude/skills/, a project's .claude/skills/, or the Settings → Capabilities → Skills upload in the Claude apps. The skill runs no scripts.
How do I install a Claude skill from a ZIP file?
Unzip the downloaded file and copy the folder into ~/.claude/skills/ for personal use or into .claude/skills/ inside a project. In the Claude apps, you can instead upload it under Settings → Capabilities → Skills.
Is this the real Paul Graham's opinion?
No. The skill says the persona is an imitation based on public essays, interviews and reported views. Its files state that the answers are inferred and not his actual view.
Can I use it for medical, legal or financial decisions?
The skill says Graham has not thought much about those areas, so its answers there would be speculation. Use a qualified professional for those decisions and treat the skill only as a way to think through general questions.
How is this different from just asking a chatbot about Paul Graham?
The skill gives the agent a fixed set of mental models, a required research step before company or event questions, and rules for the first-person voice and for exiting the persona. A general chatbot prompt usually has none of those structured steps.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
This skill suits founders, students and writers who want a sharp, essay-style sounding board on startup, product and writing questions, and who understand it is an imitation. It is a poor choice if you need verified facts without checking them, or advice on medical, legal or financial matters, where a qualified professional or a source you can verify is the better option.
Download the skill