Critical Reading

Critical Reading

Extract from a text its claims, evidence, assumptions, and omissions — reading what it does, not just what it says.

Why

Texts are optimized to persuade, and passive reading absorbs the author's frame along with their facts. Structured extraction separates what was demonstrated from what was asserted from what was implied by omission — three very different epistemic statuses fluent prose blurs together.

Use when / Don't use when

• Use when: any consequential document, especially polished, confident, or agreeable prose — agreement is exactly when critical reading is hardest and most needed.
• Don't use when: reading purely for orientation — skim first, read critically what turns out to matter.

Inputs → Outputs

• Inputs: a document — paper, memo, article, contract, transcript.
• Outputs: a claims inventory tagged by support level, the argument's skeleton, its assumptions, and its notable silences.

Principles

• Identify the conclusion first, then audit backwards — texts bury weak links in the middle.
• Distinguish demonstrated, cited, asserted, and insinuated claims as separate categories.
• What's missing is itself a claim — the comparison never made, the denominator never given.
• Note the text's frame before it sets your own.

Procedure

1. Skim for structure: what is this text trying to get the reader to believe or do?
2. Extract the main claims verbatim — early paraphrasing hides sleight of hand.
3. Tag each claim's support: demonstrated here, cited (check what the citation actually says), asserted, or implied.
4. Surface the assumptions the argument needs but never states.
5. Hunt omissions: what would a knowledgeable skeptic expect this text to address, and where is its silence?
6. Separate verdicts: what did this text actually establish versus merely say — report anything unresolved as residual uncertainty.

Common mistakes

• Critique-by-vibes — disliking a conclusion and treating that as an audit of its support.
• Trusting citations without opening them; they routinely don't say what they're cited for.
• Reading hostile texts critically while reading friendly ones passively.

Examples

• A vendor whitepaper making a competitive claim.
• A policy brief, an earnings call transcript, or a paper's methods section.
• Rereading your own draft a day later with the same scrutiny you'd apply to someone else's.

Related

• [[argument-mapping]] — formalizes the claim skeleton this skill extracts into a fully structured map.
• [[frame-detection]] — the companion check on how the text's framing pre-loads its own conclusion.
• [[statistical-audit]] — the deeper numeric follow-up whenever a text leans on quantitative claims.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: skyf0xx
• Source: skyf0xx/better-thinking
• License: MIT
• Homepage: https://zenbin.org/p/better-thinking

Install and usage instructions live in the source repository linked above.