ljg-skills

Install

Works in

• Codex
• Claude Code
• Koda
• Coddy

npx skillfoxx add skills/ljg-skills

What it does

ljg-skills is a collection of 26 skills that the author calls his set for Codex; the plugin manifest describes it as a personal collection for Claude Code. The skills fall into three groups. The first group helps you read. `ljg-paper` explains a research paper to a reader with no specialist background: what was studied, what the authors contribute, how the findings follow and what the evidence does not show. `ljg-book` breaks down a book, `ljg-read` reads a text with you, `ljg-qa` turns material into a chain of questions and answers, `ljg-map` draws a concept map, and `ljg-classic` lays a classical Chinese text out with word-by-word notes as one long image. The second group helps you think: `ljg-learn` dissects a concept from eight directions, `ljg-think` digs down to the essence of an idea, `ljg-rank` looks for the minimal set of independent causes in a field, `ljg-constraint` finds constraints and false walls, `ljg-structure` extracts a recurring motif, `ljg-analogy` picks an analogy, `ljg-mechanism` checks a phenomenon's mechanism with web search, `ljg-blind` looks for blind spots, and `ljg-roundtable` runs a round table in the names of real people. There are more, such as `ljg-explain`, `ljg-teach` and `ljg-relationship`. The third group produces output. `ljg-card` builds text into PNG in one of four molds: the long card `-l`, source-faithful full text `-f`, a comic `-c` and a whiteboard `-w`. `ljg-business-model` builds a nine-block business model canvas as PNG, HTML and JSON. `ljg-present` turns a talk into one offline HTML file, `ljg-writes` writes an article and `ljg-word` dissects an English word. Notes are saved in org-mode by default, and the Markdown version installs from the `md` branch.

Who it is for. For people who read Chinese or are fine working with Chinese instructions: students, researchers and writers who break down papers and books and turn notes into cards.

Example request

Explain this arXiv paper so that a person without a scientific background can follow it, and show what it does not prove

Limitations

The author calls the repository his personal set, not a product. Two skills are tuned to his own environment. `ljg-push` is hard-wired to the author's repository and to his local folder `~/code/ljg-skills`. `ljg-blind` reads yesterday's AI conversations from `~/.codex/sessions` and picks a chapter from WeChat Reading, taking its goal from the author's personal file.

`ljg-card` needs Bun, Playwright and Chromium, and its dependencies install with `bun install && bunx playwright install chromium`. `ljg-business-model` needs Bun and a working browser screenshot. The `-l`, `-c` and `-w` modes create raster images through your agent's own capabilities when needed.

Font names in `ljg-card -f` assume a system with KingHwa_OldSong, otherwise fallback Chinese fonts are used. The plugin manifest version is 1.17.122.

License MIT.

How to disable. Delete the `ljg-*` folders from `~/.agents/skills/` or from your agent's skills directory. Skills installed without the `-g` flag live in the project's `.agents/skills/`. Finished notes and images stay where the skill saved them and are deleted separately.

README in short

The Chinese description opens with a line about the author's set for Codex. Next come install commands through the skills CLI for the whole set, one skill and the Markdown branch, a table of flags, a separate guide for `ljg-card` and installation by cloning. The bulk is a table of 26 skills with a short purpose for each, and the end explains how the branches differ and which skills answer right in the chat. The repository holds plugin manifests, install scripts, examples, tests and evaluation cases for some skills.