Literature In Depth Reading

Literature In-Depth Reading

A reusable prompt that makes any LLM read academic papers deeply, using a first-principles, domain-expert framework. Works with ChatGPT, Claude, Gemini, local models, or Claude Code as a native plugin.

Use

Install once per platform. After that, just send the paper — no more copy-paste.

Claude Code

Inside Claude Code, run:

/plugin marketplace add XuananLu/literature-In-depth-reading-skill
/plugin install literature-in-depth-reading@xuananlu

The skill then auto-triggers whenever you hand Claude Code a paper.

Other platforms

Copy the body of skills/literature-in-depth-reading/SKILL.md (everything below the --- frontmatter) and paste it into the platform's persistent-instruction slot — once:

• Claude.ai — create a Project → paste into Project instructions.
• ChatGPT — create a Custom GPT → paste into Instructions.
• Gemini — create a Gem → paste into Gem instructions.
• Any API — put it in the system field of your requests.

From then on, every chat / call inside that Project / GPT / Gem already knows the framework. Just send the paper.

One-off try without setup: paste the prompt body as the first message in a fresh chat, then send the paper.

Supported inputs

• Pasted text
• PDF file (upload or drag in, if your platform supports it)
• arXiv or other paper URL (if your platform can fetch URLs)

Example prompts

Any language, the output matches your language.

• analyze this paper
• what's the novelty here
• break down the motivation
• 分析这篇论文

What you get

A structured analysis in six sections:

1. Task — the problem, stated formally.
2. Challenge — why it is hard and where prior methods fail.
3. Insight & Novelty — Inspiration → Insight → Novelty, with each novelty written as [problem] → [insight] → [design].
4. Potential Flaw — where the setting is too narrow, what data would break it, and which flaw is worth a follow-up paper.
5. Motivation — the author's idea reconstructed as first-principles questions.
6. Code — link to the official repo and a walkthrough of the code that implements the core novelty.

Ground rules

• Every claim about the paper is traceable to the paper.
• Outside knowledge is labeled: "Based on established knowledge…".
• If the paper does not cover something, the prompt says so instead of guessing.
• Code is only discussed when it is verifiable — from the repo or the paper itself.