---
name: "critical_thinking_workshop"
description: "Teach critical thinking through the claim-evidence-reasoning chain: how to spot weak warrants and fallacies in arguments, and how to run daily drills that build the reasoning habit. Use when the user wants to practice critical thinking, evaluate an argument, design reasoning exercises, teach reasoning, or check whether a claim actually holds up."
---

# Purpose

Most people argue with claims and evidence but never inspect the link between them. This skill trains the eye for that link. It gives the agent a working model of how arguments hold together (the claim-evidence-reasoning chain), a way to spot where they break (weak warrants and fallacies), and a catalog of drills that turn the inspection into a daily habit. Use it to run a workshop for a learner, to build exercises, or to audit your own reasoning before you commit to an answer.

# Workflow

## 1. Set up the chain

Start every engagement by teaching or applying the chain:

- **Claim**: the debatable assertion. If it cannot be disagreed with, it is not a claim, it is a fact, and there is nothing to reason about.
- **Evidence**: data, observations, or facts that support the claim. Evidence must be relevant, verifiable, and representative, not an outlier cherry-picked to fit.
- **Reasoning**: the explanation of HOW the evidence leads to the claim. This is the step most people skip; they restate the evidence instead. Force it with the prompt: "This evidence supports the claim because..."
- Full analysis adds the rest of the Toulmin model: the **warrant** (the usually-unstated assumption connecting evidence to claim), **backing** (support for the warrant), **qualifier** (how certain the claim is), and **rebuttal** (the conditions where the claim fails). See `references/argument-chain.md`.

## 2. Hunt the weak link

Most bad arguments have true evidence and a false or missing link. Diagnose:

- Name the unstated warrant: what must the arguer be assuming for the evidence to prove the claim? Write it as a sentence starting with "The argument assumes that..."
- Test the warrant with two questions: "Is the assumption actually true?" and "Is it enough, or are there other explanations the evidence fits just as well?"
- Run the fallacy check in `references/fallacy-spotter.md`. Name the fallacy if one fits; name the bare warrant failure if none does. Do not force every weakness into a named fallacy.
- Demand the rebuttal: ask what evidence would change the claim. If the claim survives all possible evidence, it is not reasoning, it is a slogan.

## 3. Run a drill

Pick a drill from `references/drills.md` matched to the user's time and setting. Rules for every drill:

- One chain per drill. Drill the whole claim-evidence-reasoning loop once, completely, rather than rattling through ten claims.
- The reasoning sentence is the deliverable. If a participant produces a claim plus evidence but no reasoning sentence, the drill is not finished.
- Close with a self-score using the rubric at the end of `references/drills.md`.

## 4. Build the habit

Critical thinking is a habit, not a lesson. End any session with a daily routine: one assumption-spotting moment (a headline, an ad, a meeting statement) and one full chain written down. Two minutes a day beats a two-hour workshop once a year.

# Output Contract

What the agent produces, depending on the mode:

- **Workshop mode**: a short chain primer, 1-2 drills run interactively with the user, and a scored rubric review at the end.
- **Exercise-design mode**: a drill plan with the question, sample evidence set, expected claim options, and the rubric, in a format the user can reuse (markdown handout).
- **Audit mode**: a claim broken into claim/evidence/warrant, the weak link named, and a verdict phrased as a qualified claim, never as absolute truth.
- Every output names its sources or assumptions. Unverifiable claims are cut or framed as opinion.

# Operating Rules

1. Every factual claim in a drill or handout must be verified by web search or a primary source. Never invent numbers, dates, quotes, or events. Unverifiable means cut it or frame as opinion. No confident unverified claims, ever.
2. Never do the reasoning for the user. Prompt, scaffold, and correct, but the reasoning sentence must come from them. This is the whole point of the skill.
3. Keep drill language plain and direct. No em dashes anywhere. No motivational-poster filler.
4. When the user states a position, drill steelmanning: state their case better than they did before you raise the objection. Attack the argument, never the person.
5. Apply the chain to your own work too: before giving a conclusion, check that your own evidence, warrant, and qualifier are in place.