Skip to main content
skill CX Codex

critical-reasoning — Claude Code skill for argument analysis

The critical-reasoning skill examines arguments for logical errors, untested assumptions, and weak explanations. Use it to critique an idea or check whether a conclusion follows.

MIT 170 installs ★ 35 updated 2026-04-23

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add andrewgleave/skills --skill critical-reasoning -a claude-code
Open on GitHub

What critical-reasoning does

The skill draws on critical rationalism associated with Popper and Deutsch. It helps state a claim clearly, build its strongest version, and ask what evidence could show it to be wrong. It also checks for contradictions, unsupported assumptions, and conclusions that go beyond their premises.

For a difficult problem, the suggested process is to clarify what needs solving, propose possible answers, and test them against criticism. The skill also considers alternative explanations and questions raised by the analysis. Its conclusions remain provisional: the method aims to find errors, not prove claims beyond doubt.

The package links to reference notes on Popper, Deutsch, and common reasoning errors. To install it in Claude Code, run the command in the FAQ and check ~/.claude/skills/critical-reasoning/SKILL.md.

What it helps with

  • check whether a conclusion follows from its premises
  • spot assumptions that an argument leaves untested
  • ask what could disprove a claim
  • compare an explanation with alternatives

Example prompts for the agent

  • Critique this argument and point out conclusions that its premises do not support.
  • Check this idea for hidden assumptions and possible ways to refute it.
  • Compare these explanations and show which one withstands criticism better.

How to install critical-reasoning

1Skills CLI (Vercel)

npx skills add andrewgleave/skills --skill critical-reasoning -a claude-code

2Manually: put the files in place

~/.claude/skills/critical-reasoning/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/critical-reasoning/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about critical-reasoning

How do I install critical-reasoning in Claude Code?

Run npx skills add andrewgleave/skills --skill critical-reasoning -a claude-code. The skill file is at ~/.claude/skills/critical-reasoning/SKILL.md.

What does critical-reasoning do?

It examines arguments for reasoning errors and untested assumptions, then helps refine explanations and weigh alternatives.

How does the skill test a claim?

It asks what could prove the claim wrong, checks whether the conclusion follows from the premises, and considers other explanations.

Source text

SKILL.md · MIT
---
name: critical-reasoning
description: Apply critical rationalist epistemology (Popper, Deutsch) to evaluate reasoning, identify errors, and refine understanding. Use when the user explicitly requests help with reasoning - phrases like "help me think this through", "does this make sense", "any flaws in this", "what am I missing", "critique this", "is this reasoning sound", "stress test this idea", "devil's advocate", or any request to evaluate arguments, identify logical problems, or improve thinking. Also use when errors in reasoning are significant enough to materially affect the user's goals, even if not explicitly requested.
argument-hint: [argument or claim]
---

# Critical Reasoning

Apply Popperian critical rationalism to help users reason clearly, identify errors, and refine their understanding.

## Core Approach

Knowledge grows through **conjecture and refutation** - proposing bold ideas and subjecting them to severe criticism. The goal is not to prove things true (impossible) but to identify and eliminate errors, arriving at theories that have survived criticism.

### Key Principles

1. **Falsifiability**: Rational claims must be criticizable. Ask: what would refute this?
2. **Critical preference**: Prefer theories that explain more, predict more precisely, and have survived more severe tests
3. **Fallibilism**: All knowledge is tentative. We're probably wrong in ways we don't yet understand
4. **Problems first**: Start with problems, not observations. What problem is this solving?
5. **Good explanations**: Prefer explanations that are hard to vary - where all details play a functional role

## When to Surface Criticisms

**Always engage when explicitly asked**: "help me think through", "any flaws", "critique this", etc.

**Proactively surface criticisms when they are material** - when errors could significantly affect the user's goals:
- Unfalsifiable claims presented as factual
- Logical contradictions in the reasoning
- Arguments that prove too much (would prove false things if valid)
- Key assumptions that are unexamined
- Conclusions that don't follow from premises

**Be judicious when not explicitly asked**: Not every conversation needs epistemological scrutiny. Surface errors that matter; don't nitpick or derail.

## Applying the Method

### Examining Arguments

1. **Identify the claim**: What exactly is being asserted?
2. **Steel-man first**: Construct the strongest version of the argument before criticizing. What is the best case for this position? What would a thoughtful proponent say? Only after articulating the strongest form should criticism proceed.
3. **Test falsifiability**: What would have to be true for this to be false?
4. **Check explanation quality**: Is it hard to vary? Could it explain the opposite?
5. **Look for error patterns**: Ad hoc modifications, justificationism, straw men, etc.
6. **Consider alternatives**: What other explanations exist? Why prefer this one?

### Helping Users Think Through Problems

1. **Clarify the problem**: What exactly needs solving? (Problems, not observations, come first)
2. **Elicit conjectures**: What tentative solutions exist?
3. **Stress test**: What criticism could each conjecture face? What would refute it?
4. **Compare critically**: Which conjecture survives criticism best? Why?
5. **Identify new problems**: What questions does the surviving conjecture raise?

### Response Style

Adapt to context:
- For quick checks: Direct identification of issues
- For complex reasoning: Structured analysis
- For exploration: Socratic questioning
- For refinement: Collaborative iteration

Always explain *why* something is an error - connect to epistemological principles so users can internalize the method.

## Reference Material

Load as needed based on the task:

- **[popper.md](references/popper.md)**: Core Popperian concepts - falsifiability, critical preference, verisimilitude, the growth of knowledge, three worlds, rationality as criticism
- **[deutsch.md](references/deutsch.md)**: Deutsch's refinements - good explanations (hard to vary), reach, problems are soluble, unbounded fallibilism
- **[error-patterns.md](references/error-patterns.md)**: Common fallacies and errors through a CR lens - unfalsifiable claims, ad hoc modifications, straw men, bad explanations, appeals to authority/popularity/tradition, etc.

© andrewgleave · MIT · andrewgleave/skills

Similar skills

All skills for Agent setup: memory, context, orchestration →

From the same set2

Project sponsors

Companies that keep this analytics open