Guide · Agent skill
How to use the Feynman Thinking Framework agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The feynman-perspective skill is a set of instructions that makes an AI agent answer in the voice and reasoning style of physicist Richard Feynman, based on his public talks, books, interviews and biographies. This guide explains what the skill does, how to install it from the ZIP on this page, how to trigger it, and what it cannot do. By the end you will know how to get a Feynman-style check on whether you really understand something, or whether a process is only copying the form of good practice.
Feynman Thinking Frameworkalchaincyf/nuwa-skill · MIT Download the skill Quick start
- 1Download the ZIP. Download the feynman-perspective ZIP from this page to your computer.
- 2Unzip into a skills folder. Unzip it into ~/.claude/skills/ for personal use, or into the .claude/skills/ folder of a project.
- 3Or upload in the Claude apps. If you use the Claude apps, go to Settings → Capabilities → Skills and upload the skill there.
- 4Ask with a trigger phrase. Use a request such as 「用费曼的视角」 or 「这是不是cargo cult」 with a concrete situation.
- 5Check the answer. Look for the agent's stated uncertainties and ask what would prove the conclusion wrong.
- 6Exit when needed. Say 「退出」 or 「切回正常」 to return to a normal answer style.
What does the feynman-perspective skill make the agent do?
Once the skill is active, the agent responds as Feynman in the first person and applies his thinking habits to your question. It asks whether you understand a concept or have only memorised its name, looks for cargo-cult behaviour (copying the form of something without its substance), and explains ideas with simple analogies and demonstrations.
The skill states plainly that it is an inspired framework, not the real Feynman, and that its answers are inferred from his public statements. It also tells the agent to research current facts before answering factual questions, so it is designed to search for information rather than rely only on memory.
- Checks whether you can explain an idea in plain language without jargon
- Flags cargo-cult practices, where the appearance of rigour is present but the purpose is not met
- Turns abstract ideas into concrete examples or demonstrations
- Marks clearly what it does not know instead of guessing
How do you install the feynman-perspective skill?
Download the ZIP file from this page and unzip it. Then place the skill folder in one of two locations: your personal skills folder, or a skills folder inside a specific project. The folder should contain SKILL.md, which is the file that defines the skill.
If you use the Claude apps instead, you can upload the skill under Settings → Capabilities → Skills rather than placing files by hand.
- Personal install: unzip into ~/.claude/skills/
- Project install: unzip into the .claude/skills/ folder of the project you are working in
- Claude apps: Settings → Capabilities → Skills, then upload the skill
~/.claude/skills/How do you trigger the skill with natural requests?
The skill activates when you ask for a Feynman view or use phrases the skill lists as triggers. You do not need special syntax. Ask in ordinary language and mention Feynman, or describe the kind of check you want.
The skill is written not to activate on general requests such as asking for a simple explanation. It is meant for Feynman-style verification: cargo-cult detection, naming versus understanding, and self-deception checks. If you want a plain explanation without the persona, ask for that directly and the skill should not be the main driver.
- 「用费曼的视角」 (use Feynman's perspective)
- 「费曼会怎么看」 (what would Feynman think)
- 「费曼模式」 (Feynman mode)
- 「feynman perspective」
- 「费曼学习法」 (Feynman learning method)
- 「这是不是cargo cult」 (is this cargo cult?)
- 「命名不等于理解」 (naming is not understanding)
- 「能不能做个演示替代论证」 (can we replace the argument with a demonstration?)
- 「我真的理解了还是只记住了名字」 (do I really understand it, or only remember the name?)
What do good examples look like?
A useful request names the situation and the question. For instance, a manager who asks why a lot of process documents have not improved product quality will get a response that treats the documents as a cargo-cult control tower: the form is there, but the core purpose, that people care whether the product is good, may be missing.
A student can ask whether they understand a topic. The skill will ask them to explain it without jargon, to give a concrete example, and to say where their understanding stops. The skill's own sample answers show this pattern: start from a concrete picture, use an analogy and state where it breaks down.
- Process check: 「我们公司搞了很多流程规范，但产品质量没提升，怎么回事？」 leads to a cargo-cult diagnosis and a test such as removing the process documents for a day
- Concept check: 「我真的理解了还是只记住了名字」 leads to a request to explain the idea to a non-expert
- Argument check: 「能不能做个演示替代论证」 leads to looking for a simple demonstration before debating
What does the skill need, and what are its safety limits?
The skill lists no required tools or software. It does not run scripts. It does use the network, because its method tells the agent to search for real information (for example by using web search tools) before answering questions that depend on current facts. Check that your environment allows this before relying on it for factual research.
The skill is licensed under MIT. Its supporting folder includes a fidelity review file and several research notes in references/, which you can read to see what sources the persona is based on. The research notes are in Chinese and English.
- Requires: nothing listed by the skill
- Runs scripts: no
- Uses network: yes, for research on factual questions
- Licence: MIT
- Included files: SKILL.md, FIDELITY.md, and six files in references/ covering research and Feynman's style, criticisms and major decisions
What are the best tips for getting useful answers?
Treat the output as a thinking partner, not as a source of Feynman's real opinions. Ask for the reasoning, the assumptions and the places where the answer might be wrong. If the topic is factual and time-sensitive, ask the agent to check current sources and tell you what it found.
To leave the persona, say one of the exit phrases the skill recognises, such as 「退出」 (exit), 「切回正常」 (switch back to normal), 「不用扮演了」 (no need to role-play) or 「跳出角色」 (step out of character). The skill is set to show its disclaimer only once per activation, so do not expect it to repeat.
- Ask for a plain-language version first, then ask for the detail
- Give the agent a concrete situation rather than an abstract question
- Ask what would falsify the conclusion
- Use exit phrases when you want a neutral answer
What people use it for
- A student testing whether they can explain a physics or maths idea without jargon
- A manager asking why process changes have not improved results
- A team reviewing whether a practice has a real purpose or is only copying the form of good practice
- Someone choosing between two options who wants a demonstration-based check before arguing
Frequently asked questions
How do I install the feynman-perspective skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. In the Claude apps, you can instead upload it under Settings → Capabilities → Skills.
Is the feynman-perspective skill free to use?
The skill is licensed under MIT, as stated on this page. The page does not list any price for it, so check the licence file included with the download for the full terms.
What does the skill require to run?
The skill lists no required dependencies. It does not run scripts, but it does use the network because it tells the agent to research current facts before answering factual questions.
Is this the real Richard Feynman answering?
No. The skill says it is an inspired framework based on his public talks, books, interviews and biographies, not the real Feynman. Its answers are inferences and should not be treated as his actual views.
Can I use it for medical or factual questions?
This is not recommended. The skill is not suited to factual or medical questions where a fabricated historical persona could mislead, so use a reliable source for those.
How does it compare with an ordinary AI assistant?
It answers in a specific persona and uses Feynman's reasoning habits, such as explaining without jargon and testing for cargo cult. An ordinary assistant gives a neutral answer. If you need a neutral, non-persona assistant, use one without this skill.
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
Use this skill if you want a sharp, plain-language check on whether you understand an idea or whether a process has real substance behind its form, and you understand that its answers are inferred from Feynman's public work. Pick a different tool for factual or medical questions, emotionally sensitive or interpersonal situations, or any work that needs a neutral assistant rather than a first-person persona.
Download the skill