ljg-skill-xray-paper: The Paper Explainer That Refuses to Be Neutral

A Claude Code Skill that loads `soul.md` and `memory.md`, strips jargon to the bone, and turns academic papers into a personalized x-ray in Org-mode and ASCII.

8 min read · lijigang/ljg-skill-xray-paper

An academic paper is split by a sharp visual seam into two worlds. One side feels dense and abstract, the other side feels like a reader's notes, cards, and threads reaching back into the text. It explains that this skill compares a paper with stored context instead of compressing it in isolation. — The repo's core move is not summary. It is friction.

论文 X 光机 — 一个 Claude Code Skill，只做两件事：论文说了什么 + 对我意味着什么。

lijigang, Project Creator · lijigang/ljg-skill-xray-paper

Key Takeaways

ljg-skill-xray-paper treats paper reading as a collision between a paper's claims and a stored worldview, not as a neutral summary task.
Loading `soul.md` and `memory.md` turns Claude Code into a personalized reasoning engine that shows what changed in the reader's mind.
The repo's zero-jargon ladder and Org-mode output turn explanation into a disciplined progression from scene to mechanism to terminology.
Its real edge is not polish, but a terminal-native workflow that keeps analysis portable, inspectable, and opinionated.

Most paper tools answer a familiar question: what does this paper say? `ljg-skill-xray-paper` asks a stranger, better one: what does it say relative to the rest of your brain? The output is not a neutral digest. It is a structured delta between a paper and a remembered worldview.

The strange power of `soul.md` and `memory.md`

The repository's most interesting move is simple. Before analysis begins, the skill reads `soul.md` and `memory.md`. That gives the paper a baseline to bump into. The result is not just what the paper says, but where it updates, contradicts, or sharpens the reader's prior model.

A close-up desk scene shows a notebook, a stack of index cards, and a paper printout pulled side by side. One card is pinned beside a claim with a thin thread of tension between them. It explains how the skill turns memory into an active participant in reading. — The skill treats memory files as active participants in the reading process.

零术语规则：承重概念必须场景化（3 级渐进），去掉技术名字仍能理解

lijigang, Project Creator · lijigang/ljg-skill-xray-paper

How the x-ray machine works

The pipeline is deliberately plain. Claude Code normalizes arXiv links to HTML, because structured HTML is easier to scrape than raw PDF layout. Then the skill extracts load-bearing concepts, explains them through a zero-jargon ladder, and emits an Org-mode report with ASCII sketches that still make sense in a terminal.

The paper does not go straight to summary. It crosses memory first, then becomes a report.

The interactive version matters because the logic is sequential. Hover a stage and the rest should dim. Click the collision node and the reader should see why that claim mattered. The point is to make the join between paper and memory feel like the central event, not a hidden implementation detail.

Zero jargon, but not zero depth

A staircase rises from an ordinary scene into a more abstract machine. The lower steps are grounded in everyday objects, while the top step suggests a technical concept arriving only after the climb. It shows that the skill delays jargon until the explanation has already earned it. — The repo delays technical terms until the concept has already been explained in ordinary language.

This is the skill's best editorial rule. It does not ban technical terms. It makes them earn their place. A concept should survive a plain-language scene, then a mechanism, then the technical label. That keeps explanation from collapsing into keyword worship.

* What the paper said
** Elevator pitch
** Load-bearing concepts
** Napkin sketch

* What it means to me
** Cognitive collisions
** Updated mental model

一句话压缩：电梯里跟外行朋友说的那句大白话 + 餐巾纸图

lijigang, Project Creator · lijigang/ljg-skill-xray-paper

What it beats, and what it doesn't

The comparison is less about features than about epistemology. ChatGPT and Claude can summarize a paper. ChatPDF-style tools can extract it. Elicit and Scholarcy can organize research work. This repo is doing something narrower and stranger: it turns reading into a confrontation with context.

Tool	What it gives you	What it misses	Best use
Generic LLM chat	Flexible answers from pasted text	Context is ad hoc and the output drifts toward summary	Quick questions and loose exploration
ChatPDF-style summarizer	Fast extraction from a document	Mostly compresses the paper itself	Skimming a single paper
Elicit or Scholarcy	Structured research workflows and literature helpers	Built around their own product flow, not your memory files	Broader literature review work
ljg-skill-xray-paper	An Org-mode x-ray against `soul.md` and `memory.md`	Requires a Claude Code workflow and a user who wants friction, not polish	Close reading that updates a worldview

That is why the repo stands out. It is not trying to win the category of paper summarizers. It is trying to build a better habit. The habit is: read, compare, and then decide what changed in your own model.

Why this matters in Claude Code

This is a clean example of where terminal-native AI tools are headed. The skill is tiny, opinionated, and text-first. It sits beside code instead of replacing your workflow. That makes it useful in a way a glossy app cannot always match: the output stays portable, inspectable, and easy to remix.

For founders and product people, the lesson is sharper. You do not need a bigger interface to make an AI tool valuable. You need a stronger theory of use. `ljg-skill-xray-paper` has one.