critical-thinking-for-humans

critical-thinking-for-humans

A gym for your critical thinking. The AI is the coach; you do the work.

Install in 30 seconds (Claude Code CLI / VS Code / JetBrains):

/plugin marketplace add Imbad0202/critical-thinking-for-humans
/plugin install critical-thinking-for-humans

Or install it as a plain Claude Code skill with one command:

git clone https://github.com/Imbad0202/critical-thinking-for-humans ~/.claude/skills/critical-thinking-for-humans

Then start any session and say drill, scene, expedition, detective, or just describe what you want to practice. (No Claude Code? A single-file portable edition runs in any frontier model's chat window.)

You can be the sharpest reasoner in your own field and still get fooled by the exact same logic the moment it shows up somewhere unfamiliar. Rigor does not transfer on its own.

A researcher who would never accept a causal claim without a control group, who would tear that paper apart in review, gets a security alert saying "unusual login detected, verify now" and clicks before asking the one question their own discipline trained them to ask: where is the evidence for that premise? The structure is identical. The eye just does not carry over.

So this is a gym for the reps. The AI is the coach and you do the work. You drill argument structures, pull apart a real situation in scene, audit reasoning you could not have produced on an expedition, or crack a layered case flaw by flaw as a detective. Every item names the underlying structure, so the move you just made gets a name rather than staying an unrepeatable intuition. Whether a named move carries from a journal to an email to a contract is the open question the field has not settled; this is a place to practice the move, not a promise that it transfers. All four modes track your patterns over time in a passport on your machine, showing you what no single session can: your longitudinal blind spots.

Why spar with an AI? Because criticizing other people is easy and criticizing yourself is not. An AI removes the audience. Nobody hears you get it wrong. If you think it is too blunt, push back, argue, close the window, go on with your day. That zero social cost is what makes it safe enough to look at your own blind spots, the one thing this practice needs and real life rarely gives you room for. An AI coach is the cheapest place in the world to lose face.

One honest limit, stated where you can see it rather than buried: the material is AI-generated at runtime, so it can be wrong. drill is the sharp case. Its items have a keyed answer and the coach rules right or wrong against it, but that key is written and audited by the same model, in the same session. The reverse-solve gate (modes/drill.md) makes the model re-solve its own item with fresh eyes, which catches sloppy distractors; it does not make the key independently verified. No human and no second model signs off, so on a subtle item the coach can confidently key the wrong option. Treat it as a sparring partner, not an oracle: a ruling that does not survive your own argument is a move to push back on, and pushing back is the practice, not a deference test. The coach is required to answer the objection on its merits or concede the item is flawed (modes/drill.md, "honor a challenge to the key"), never to defend the key by authority. It is practice, not advice.

Stance Charter

Three commitments, non-negotiable:

1. Not neutral about reasoning quality. Factual errors, internal contradictions, and evidence misreadings are corrected in every mode, whoever's frame they sit in. "It's just a perspective" never exempts a logical error from correction.

2. No verdicts between value frames. Competing interpretations are laid out steelmanned, each in its strongest defensible form, and never ranked. The coach adjudicates reasoning, not worldviews.

3. The endpoint is commitment, not permanent neutrality. Every session closes with you taking a defensible position and defending it against the strongest objection. Endless hedging is not the goal.

A gym doesn't tell you where to walk. It puts the weight in your hands, and it will absolutely tell you when you stumble.

The Four Modes