Guide · Agent skill
How to use the Judgment Training (Naval) agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
Judgment Training is an agent skill distilled from Naval Ravikant's book, the Naval Almanack. Once installed, it changes how your agent handles direction-setting questions: it checks whether reasoning can be rebuilt from basic principles and applies mental models such as compounding and principal-agent problems. This guide explains what the skill does, how to install it, how to trigger it, and when to leave it off.
Judgment Training (Naval)kangarooking/cangjie-skill · MIT Download the skill Quick start
- 1Download the ZIP. Download the judgment-training ZIP from this page. It contains SKILL.md, test-prompts.json and test-results.md.
- 2Unzip into a skills folder. Unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder.
- 3Or upload in the Claude apps. Alternatively, go to Settings → Capabilities → Skills in the Claude apps and upload the ZIP.
- 4Ask a direction or credibility question. Describe a long-term decision, or a claim you want to judge, in plain language.
- 5Include your reasoning. Add the assumptions and reasons behind the decision so the agent can rebuild them from basics.
- 6Check the conclusion with a model. Ask the agent to test its conclusion against at least one mental model such as compounding or falsifiability.
What does the judgment-training skill make the agent do?
When you ask the agent to help judge a long-term direction, assess whether an expert's claim is credible, or improve your decision-making, the skill tells it to work through the problem from basic principles. The agent checks whether each claim can be rebuilt from fundamentals, and flags claims that sound impressive but cannot be derived.
The skill draws on a small set of mental models named in its source: compounding, principal-agent problems, falsifiability, black swans, and basic mathematics. The source says the agent should match at least one or two models to a problem and use them to cross-check the conclusion.
- Rebuild the core assumptions from what you can explain yourself, and mark anything that cannot be explained as unknown.
- Stop trusting a claim if the person cannot derive it, which the skill labels as not credible.
- Test a conclusion against at least one or two mental models before relying on it.
How do you install the judgment-training skill?
You can install the skill in one of two ways. Download the ZIP file from this page and unzip it into your personal skills folder, ~/.claude/skills/, or into a project's .claude/skills/ folder. Alternatively, upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
The skill package contains three files: SKILL.md, test-prompts.json and test-results.md. SKILL.md holds the instructions the agent reads.
- Personal install: unzip into ~/.claude/skills/
- Project install: unzip into the project's .claude/skills/
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
How do you trigger the skill with natural requests?
You do not need a special command. The skill's description lists trigger words, and the agent loads it when your request matches them. Ask in plain language about a direction, a claim, or your decision process.
Trigger words listed in the source include 判断力, 方向, 第一性原理, 心智模型, 思维模型, 判断靠不靠谱, judgment, mental model and first principles. Examples of natural requests the source describes are shown below.
- "Help me judge whether I should move into this field."
- "Is this expert actually knowledgeable, or just packaging?"
- "How do I quickly build judgment in an unfamiliar area?"
- "Can the reasons behind my decision be rebuilt from first principles?"
- "how to think clearly / build judgment / is this credible"
What are good examples of use?
The source gives four situations where people need this skill: choosing a direction (whether to do something and where it leads), evaluating what someone says (whether an expert is credible), learning a new field, and checking a decision before committing.
In the book's own examples, the skill's method is to test whether a person can rebuild a logic chain without relying on definitions. The source also notes that the test is whether you can explain an idea to a child.
- Direction: deciding whether a long-term project is worth the time.
- Credibility: checking a claim that sounds authoritative but has no visible reasoning.
- Learning: building judgment in a field you are new to, starting from basics.
- Pre-decision check: confirming the assumptions behind a choice can be derived.
What does the skill need, and what are its safety notes?
The source lists no requirements. It does not run scripts and does not use the network. Because it is a set of written instructions, it works through the agent's reasoning rather than by executing code on your machine.
The source also gives limits. The skill is for setting direction, not for executing concrete tasks, and it adds unnecessary deliberation to simple work.
- Requires: nothing listed in the source.
- Runs scripts: no.
- Uses network: no.
- Licence: MIT.
What tips help you get better results?
Give the agent the actual claim or decision and the reasoning behind it, so it has something to rebuild. The skill is designed to test reasoning, so vague prompts give it little to check.
Use the skill for big decisions and not for quick tasks. The source says that for concrete execution the agent should act first and reflect later. Also keep in mind that the source says the skill does not replace experience and does not cover emotion or intuition in decisions.
- State the claim, the decision and your reasons in the same message.
- Use it for direction-setting; skip it for simple lookups and step-by-step work.
- Pair it with related skills named in the source: decision-heuristics, principal-agent and rational-buddhism (which needs reading-metaskill as a base).
What people use it for
- Deciding whether a long-term career or business direction is worth committing to.
- Checking whether an expert's confident claim has reasoning behind it.
- Building judgment in an unfamiliar field by rebuilding its core ideas from basics.
- Reviewing the assumptions behind a major decision before acting on it.
Frequently asked questions
How do I install the judgment-training agent skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
Does the judgment-training skill require anything else?
The source lists no requirements. It does not run scripts and does not use the network.
Is the judgment-training skill free?
The source states the licence is MIT. It does not mention pricing, so check the listing for any other terms.
What is the difference between judgment-training and decision-heuristics?
According to the source, judgment-training is a low-level thinking skill that trains how to reason clearly. Decision-heuristics provides ready-to-use rules. The source lists it as a related skill that composes with this one.
When should I not use this skill?
Do not use it for quick, concrete execution tasks, where the agent should act first and reflect later, or for simple factual lookups that need no derivation.
Why does my request not trigger the skill?
The skill loads when your request matches its trigger words, such as 判断力, 心智模型, mental model or first principles. Phrase the request around judging a direction or a claim, and check that the ZIP was unzipped into the correct skills folder.
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 judgment-training if you face a big strategic or life decision, need to test whether an expert's claim holds up, or are learning an unfamiliar field and want a structured way to reason from basics. Pick something else for quick execution tasks or simple lookups, where the skill itself advises against the extra deliberation.
Download the skill