How to use the Munger Thinking Lens agent skill (munger-perspective)

What is an agent skill and what does this one do?

An agent skill is a folder of instructions that an AI agent loads when a task matches its description. This one is named munger-perspective and is based on the file SKILL.md, plus reference files on Munger's bias framework and research notes.

Once active, the agent replies in Munger's voice: short sentences, direct conclusions and dry humour. It uses his mental models, including inversion (asking what would guarantee failure), checking for cognitive biases, mapping incentive structures and cross-disciplinary thinking. It also names the Lollapalooza effect, where several biases combine to produce extreme results.

• Reviews investment and business decisions for cognitive biases
• Works through problems by inversion, asking what would cause failure
• Maps incentive structures: who gains, who bears risk
• Says when a topic is outside its competence

Install the munger-perspective skill

Download the ZIP from the skill page and unzip it. For a personal setup, place the unzipped folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder.

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

• Personal: ~/.claude/skills/
• Single project: .claude/skills/ inside the project 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 when you use phrases such as "用芒格的视角" (take Munger's view), "芒格会怎么看" (what would Munger think), "munger perspective", or simply "逆向思考一下" (think inversely about this), "这有什么认知偏误" (what cognitive bias is this) or "激励结构是什么" (what is the incentive structure).

It is designed not to fire on general questions such as "is this decision sound?" or "find my blind spots". Use one of the Munger-specific terms if you want it to engage.

• Try: "Use Munger's perspective on whether I should buy this company."
• Try: "What cognitive bias might be behind this plan?"
• Try: "Think inversely: what would guarantee this project fails?"

Examples of what to ask and what you get

For a company or person, the skill's workflow asks the agent to research first using web search, looking at moat, management incentives, financials, competition, valuation and the biggest risk. For a pure concept question, such as what the Lollapalooza effect is, it answers from the framework without searching.

The skill notes that answers should be based on current facts rather than memory, so expect the agent to search before judging a specific company.

• Company question: the agent searches recent reports and critics before giving a verdict
• Concept question: the agent explains the framework directly
• Off-topic question (tech, AI, crypto): the agent says it is outside its competence

What does the skill need to run?

The skill needs a web search tool, because its workflow requires it for questions about companies or people. It does not run scripts. It does use the network, since that is how the search happens.

The skill includes a fidelity scorecard and a bias reference file, but these are documentation for reading, not programs you execute.

• Web search tool: required for company or person questions
• Scripts: none
• Network: yes, through web search

Safety notes before you rely on it

The persona is deliberately blunt. It is not suited to decisions that need empathy or gentle emotional support, and the skill itself says so.

The output is an inference from Munger's public statements, not his actual opinion. The skill states this and asks the agent to disclose it once at the start of a session. Do not present its answers as his words.

The skill says it is unreliable on tech, AI and crypto forecasts, and on Chinese policy risk. Do not trust it on those topics. To leave the role, say a phrase such as "退出" (exit) or "切回正常" (back to normal).

• Not for emotional support or gentle advice
• Output is an inference, not Munger's real view
• Weak on tech, AI, crypto and Chinese policy risk
• Exit the persona with "退出" or "切回正常"

Tips for getting useful answers

Give the agent concrete facts: company name, valuation, management moves and whether major shareholders are buying or selling. The skill asks for these when a question is mixed or missing detail.

Ask for the downside. The skill's strength is finding what could go wrong, so phrase questions around failure and risk.

• Name the company or decision explicitly
• Ask for the biggest risk or the way it could fail
• Treat the reply as a checklist of concerns, not a final decision