Critical Thinking

Critical Thinking Skill for Claude

A Claude skill that switches Claude from collaborative assistant into an uncomfortable critical reviewer at the end of a working session — before you act on, publish, or share what the session produced.

What this skill does

After a substantive session with Claude (writing a post, building an analysis, drafting a report, producing an infographic), you ask Claude to run a critical review. The skill forces Claude into a different role: reviewer, not collaborator. It walks through six dimensions:

1. Starting premises you never challenged
2. The moment you shifted from understanding to producing
3. Claims and numbers accepted without verification
4. Perspectives that never entered the room
5. Complexity Claude smoothed over
6. Confidence calibration — was your trust in the output too high?

It then gives you an ownership test, a to-do list of what to verify, and a pattern-recognition note about how you worked with Claude in this session. It ends with a one-sentence verdict: ready to use, needs verification, or should not be used as-is.

Critically: the skill does not soften under pushback, and it never ends with encouragement.

Why this skill exists

Research from 2024–2026 shows that frequent AI use correlates with reduced critical thinking, mediated by cognitive offloading (Gerlich 2025). Microsoft and Carnegie Mellon found that higher confidence in AI leads to less critical engagement with its outputs (Lee et al. 2025). And the "metacognitive laziness" research (Fan et al. 2024, MIT Media Lab 2025) shows users can produce good outputs while retaining less of the material and losing the ability to defend it.

The skill is a structured intervention against those patterns. It is friction you missed during the session, delivered at the moment before you publish.

Full research citations and caveats are in `references/research-foundation.md`.

---

How to install

You do not need to know how to code