Guide · Agent skill
How to use the ljg-qa Idea Extractor agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
ljg-qa is an agent skill that reads an article, paper or book and rewrites its core reasoning as a chain of pointed questions, each with a structured answer. It is for readers who want to retrace an author's argument rather than skim a summary. This guide explains what the skill does, how to install it, how to trigger it, and what to watch for before you send it sensitive material.
ljg-qa Idea Extractorlijigang/ljg-skills · MIT Download the skill Quick start
- 1Download the ZIP. Download the ljg-qa ZIP from this page.
- 2Unzip into a skills folder. Place the ljg-qa folder in ~/.claude/skills/ for personal use, or in a project's .claude/skills/ folder.
- 3Make sure the notes folder exists. Create ~/Documents/notes/ if it is not already there, because the skill writes its output there.
- 4Start a new session. Open a fresh agent session so the skill loads.
- 5Share a source and trigger the skill. Send a URL, file path or text with /ljg-qa or a trigger word such as 问答 or Q&A.
- 6Review the note. Open the generated Markdown file and read the question chain in order, checking each boundary against the source.
What is an agent skill, and what does ljg-qa make the agent do?
An agent skill is a folder containing a SKILL.md file and supporting documents that an AI agent loads when a request matches the skill's description. ljg-qa tells the agent to find the author's core claim, identify the key turning points in the argument, and turn them into a sequence of questions where each question follows from the answer before it.
Each answer follows four parts: a one-sentence conclusion, a compact symbolic summary of the relationship, the reasoning step that leads to it, and the conditions under which it fails. The output is a Markdown note saved to ~/Documents/notes/.
- Questions focus on why an approach works, how it differs from alternatives, its costs and limits.
- The skill explicitly rejects FAQs, glossaries, comprehension quizzes and plain summaries.
- The note is named in denote style: a timestamp, then --qa- and a short topic, ending in __qa.md.
Install ljg-qa: download the ZIP and unzip it into a skills folder
Download the ljg-qa ZIP from this page. Unzip it so that the skill folder contains SKILL.md, Workflows/Extract.md and References/QuestionDesign.md.
For a personal install, place the folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder instead. Start a new agent session afterwards so the skill is picked up.
If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/ljg-qa/
- Project only: <your-project>/.claude/skills/ljg-qa/
- Claude apps: Settings → Capabilities → Skills, then upload
Trigger ljg-qa with a natural request
The skill loads when you use one of its trigger words or share a source and ask for Q&A extraction. The source lists these triggers: 问答, Q&A, QA, 提问, 抽取问题 and /ljg-qa. You can paste a URL, give a local PDF path, or paste the text directly.
The skill chooses how to fetch the content based on the input. A normal web page goes through a web fetch, an arXiv link is read through its HTML version, a local PDF or file is read directly, and pasted text skips the fetch step.
- Slash form: /ljg-qa https://example.com/article
- Plain request: Turn the main ideas of this paper into Q&A: ~/Downloads/paper.pdf
- Pasted text: 把这段抽成 Q-A: [your text]
/ljg-qa https://example.com/articleRead the output: questions, four-part answers and why the order matters
The questions are ordered by argument dependency, not by where they appear in the source. Question 2 should only make sense after Question 1 is answered, so reading the chain in order walks you through the author's reasoning.
The skill asks for a mix of question types: action (how did they do it), comparison (why A and not B), cause (why does this solution hold) and boundary (when does it fail). Using only one type produces a tutorial, a debate piece, a paper-style argument or a reflection, so the mix is what keeps the tension of the original argument.
- Symbolic summaries are plain text relations such as A = B + C or old: X → new: Y, not mathematical formulas.
- Each answer has a conclusion, a symbolic line, the reasoning steps and a boundary.
- Use the boundary parts to check whether the author's claim holds in your own context.
Get better results: give clean sources and check the boundaries
The skill performs best on dense, argument-heavy material. For long PDFs the source notes that the agent should read in sections using the pages parameter, so tell the agent which part matters most if the document is long.
Because the answers are meant to be checked against the source, read the boundary lines critically. If an answer sounds like a definition, the skill's own guidance says to rewrite it, so ask the agent to redo that question.
- Paste the full text or a direct link rather than a short excerpt.
- For pages behind a login, the workflow points to a markdown-proxy step, so check that your setup can reach the content.
- Ask for a rewrite if a question is answerable with one definition.
Safety: scripts, network access and private material
ljg-qa runs scripts and uses the network. Its workflow sends a notification with curl to a local service at localhost:31337 and suppresses any error if that service is not running. The skill also fetches URLs and reads files, so the content you share is processed by the agent.
Do not send sources you are not legally or privately allowed to share with an agent. Review the SKILL.md and workflow files before installing if you need to confirm exactly what runs on your machine.
- Output is written to ~/Documents/notes/, so that folder must exist and be writable.
- The optional notification call goes to localhost:31337 and its failure is suppressed.
- Web fetch or file reading capability must be available to the agent for URLs and PDFs.
curl -s -X POST http://localhost:31337/notify \
-H "Content-Type: application/json" \
-d '{"message": "Running Extract in ljg-qa"}' \
> /dev/null 2>&1 &What people use it for
- Working through a dense research paper section by section before a reading group.
- Building a reasoning-based note in a Markdown or denote-style notes folder.
- Checking whether an author's method holds in your own setting by reading its boundary conditions.
- Understanding how a book's argument is built before deciding whether to read it in full.
Frequently asked questions
Is ljg-qa free to use?
The source identifies the skill's licence as MIT. It does not state any price, so check the repository for any other terms before relying on it.
What do I need to run ljg-qa?
You need an agent that can load skills, a local notes folder at ~/Documents/notes/, and web fetch or file reading capability for URLs and PDFs. The skill also runs scripts and uses the network.
How do I install the ljg-qa skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or a project's .claude/skills/ folder. In the Claude apps you can upload it under Settings → Capabilities → Skills.
What trigger words start ljg-qa?
Use /ljg-qa, or say 问答, Q&A, QA, 提问 or 抽取问题, or share an article, paper or book and ask for Q-A extraction. The skill is not meant for FAQs, glossaries or quizzes.
Why does my output look like a summary or a FAQ?
The skill is built to reject those formats, so a summary-like result usually means the request was too vague. Ask again for the core argument as a chain of why, how and boundary questions, with a four-part answer for each.
How is ljg-qa different from a summarizer?
A summary restates conclusions, while ljg-qa asks questions in dependency order and answers each with a conclusion, a symbolic relation, a reasoning step and a boundary. The output is longer and more demanding than a recap.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use ljg-qa if you want to retrace the reasoning of a dense article, paper or book in a Markdown notes folder, and you are comfortable with an agent that runs scripts and uses the network. Pick a summarizer if you only need a quick recap, or a different tool if you need study quizzes, FAQs or glossaries, which this skill explicitly refuses to produce.
Download the skill