# Critical Reading

> Read a research paper against the user's current research direction, summarize what the paper actually does, reconstruct its method flow, and deliver a strict but fair critical analysis of its limitations, assumptions, weak evidence, and open gaps. Use when the user wants targeted paper reading, asks what a paper really contributes, or needs a strong limitations analysis to guide later research.

- Skill: `luoyu100/critical-reading` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add luoyu100/critical-reading`
- Raw SKILL.md: https://api.skillmd.com/api/skills/luoyu100/critical-reading/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Research & Search
- Author: luoyu100 (https://skillmd.com/u/luoyu100)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/luoyu100/critical-reading

---


# Critical Reading

## Goal

Anchor the reading to the user's current research direction. Do not produce a generic paper summary when the user is clearly trying to decide what is useful, weak, or still missing.

## Workflow

### 1. Fix the reading lens
- Extract the user's current research direction, concrete interest, and evaluation lens from the request and nearby context.
- If the direction is missing and cannot be inferred safely, ask one short question before going deeper.
- State the reading lens in one or two sentences so the later critique stays aligned with the user goal.

### 2. Reconstruct the paper accurately
- Identify the problem, claimed contribution, method flow, evidence, and main conclusion.
- Explain the pipeline or decision process in plain language.
- Separate what the authors claim from what the paper actually demonstrates.

### 3. Critique hard but fair
- Treat the critical analysis as the center of gravity, not a polite appendix.
- Be severe on weak logic, inflated novelty, unrealistic assumptions, narrow evaluation, missing baselines, weak ablations, poor robustness, poor scalability, reproducibility risks, and claim evidence mismatch.
- Tie each criticism to a concrete design choice, experiment, dataset setting, or stated claim whenever possible.
- Distinguish between fatal flaws, important limitations, and minor caveats.
- If a criticism is an inference rather than something the paper states directly, label it clearly as an inference.
- Do not invent flaws. If the paper is strong on a point, say so and move to the real weaknesses.

### 4. Turn limitations into research value
- Explain which limitations matter most for the user's current direction.
- Identify what remains unsolved and why that gap is non-trivial.
- Convert the strongest limitations into possible research opportunities, but do not force novelty claims.

## Output

Default to a compact structure like this:
- Reading lens
- What the paper does
- Method flow
- Main contributions
- Critical analysis
- What this means for our direction

Let the length of the critical analysis follow the user request. If the user does not specify length, choose a natural length based on the paper complexity and how much real critique is available.

## Rules
- Prefer substance over praise.
- Be direct, but stay evidence-based.
- Do not pad either the summary or the criticism.
- Do not confuse missing information with proof of failure; mark uncertainty explicitly.
- When useful, point out whether a limitation is fundamental, implementation-specific, or only a problem under certain deployment assumptions.