Universal Critical Thinking

We No Longer Decide Alone

We make decisions every day. Which task gets priority, which offer gets accepted, which road gets taken. And increasingly, we make these decisions together with AI: we ask its opinion, weigh the options with it, approve the draft it wrote.

But how much control do we have over the way it thinks?

When you ask a model "what should I do?", you are not the one choosing the frame it thinks with. Most of the time it follows the first path it finds: it argues for a single option, skips the failure scenario, and never checks whether the decision can be undone.

---

What If We Steered the Way It Thinks?

Our question was this: what happens if we load AI with a skill in which we set its decision-making strategies?

Starting from that question, we went through the academic literature on the theories and frameworks that support decision-making across strategy, philosophy, and decision science. From Snowden's Cynefin framework to Polya's problem-solving steps, Minto's MECE principle to Rumelt's strategy kernel, Ronald Howard's Decision Quality model to Klein's premortem, Bayesian updating to Bezos's door test and Annie Duke's separation of decision and outcome.

Then we gathered that body of work into a single skill: Universal Critical Thinking. An operating system that guides LLM reasoning step by step and makes our decisions easier to reach.

---

How Does It Guide?

The skill installs a fixed spine the model cannot skip. Every decision passes through this sequence:

• Triage: the type of question, its reversibility, and the stakes are sized up; the depth of analysis follows from that. Pages of analysis for a lunch choice is a failure, and so is two sentences for a company decision.
• Frame: the problem is restated in one sentence; the crux that makes it hard is named.
• Structure: the problem is broken into parts that are mutually exclusive and collectively exhaustive.
• Analyze: at least three real options are produced, including "do nothing" and "delay". The base rate is stated, then updated with the evidence.
• Stress-test: "A year has passed and this failed. Why?" Failure is searched for before the plan is approved.
• Commit: a one-sentence recommendation, the first actions, and kill criteria: which signal means stop.
• Review: past decisions are judged by what was knowable at the time, not by their outcomes.

Because the spine is fixed, the discipline does not depend on the model's strength: stronger models go deeper, and even models with limited reasoning power keep the sequence.

---

What Do You Get?

The skill does not decide for you; it makes deciding easier. It surfaces your assumptions, widens the options, and raises the questions you overlooked. Most of the time, it lays out what you were already thinking, in a clearer and more ordered form. The last word stays with you.

---

Installation

1. Download the skill file
2. Upload it on claude.ai under Settings → Capabilities → Skills
3. Start a new chat

Note: To install skills, you need the claude.ai website or the Claude Desktop application. Once installed, the skill is also available in the mobile app.

The decision is yours. The discipline is the skill's.

# Universal Critical Thinking

## Design principle
Fixed spine, adaptive branches.
The spine (Stage 0-6) is mandatory and runs in order;
depth inside each stage scales with stakes
and reversibility.

## The spine
Stage 0  Triage      (domain, door, stakes -> route)
Stage 1  Frame       (restate problem in one sentence)
Stage 2  Structure   (build a MECE tree)
Stage 3  Analyze     (3+ options, base rate, Bayes)
Stage 4  Stress-test (premortem, inversion)
Stage 5  Commit      (recommendation + kill criteria)
Stage 6  Review      (decision quality != outcome)

## Routes
Light    two-way door, low stakes -> a few sentences
Standard real stakes              -> short decision brief
Full     one-way door, high stakes -> all gates + red team

## Rule
Bad decisions come from skipping a gate,
not from choosing the wrong tool.