lijigang/ljg-skills
21 skills64.4k installs116k starsGitHub
Install
npx skills add https://github.com/lijigang/ljg-skillsSkills in this repo
1Ljg Cardljg-card (铸) is a generative visual skill for solo builders and creators who need social-ready graphics from existing text without hiring a designer for every post. You invoke it when a draft, article URL, or note should become a long reading card, adaptive infograph, multi-slide reading set, editorial sketchnote arc (problem through naming), manga strip, whiteboard diagram, or high-contrast big-font attachment tuned for 小红书. The workflow fetches or reads source material, picks a mold flag, renders HTML, and captures PNGs through a bundled Node screenshot script with predictable 1080px widths. That fits indie operators who publish in Chinese and English communities and want one repeatable “cast to image” step inside their agent session. It deliberately outputs binary visuals rather than Org-mode or ASCII-only docs. Use it when distribution depends on scroll-stopping visuals; skip it when you only need plain markdown in a repo or formal PDF documentation.3.8kinstalls2Ljg Roundtable圆桌讨论 (ljg-roundtable) is a journey-wide agent skill by 李继刚 that implements a structured seminar for solo builders who want depth without a real advisory board. A moderator persona anchors the session, invites representative figures aligned to different intellectual stances, and drives dialectical rounds on a user-supplied topic until contradictions surface and a knowledge network crystallizes. Each round can produce ASCII visual frames of the evolving argument map. It fits whenever you must explore positioning, ethics, architecture tradeoffs, or market narratives before locking a spec—not only at raw ideation. Intermediate complexity: you supply a clear question and stay engaged across rounds. It complements brainstorming and planning skills by adding adversarial, multi-voice reasoning rather than a single assistant monologue.3.7kinstalls3Ljg Paperljg-paper is a journey-wide agent skill that teaches agents to read research papers for solo builders who are not academic insiders. It outputs structured Org-mode notes in Chinese narrative form: stake the problem with a concrete protagonist, unfold the method on that same example, distill three indispensable concepts, extract one portable insight, run a frank advisor-style review with a verdict, and close with personal 启发 aimed at action. Canonical placement is Idea/research because that is where paper digestion usually starts, but the workflow fits anytime you must onboard a new technique before Validate scoping, Build choices, or Grow content strategy. The template enforces continuity—no orphan summaries—and keeps hype out of the 洞见 section unless the paper truly earns a takeaway sentence that stands alone.3.7kinstalls4Ljg Travelljg-travel is an agent skill workflow for travelers—not typical indie SaaS shipping—who want archaeological-style desk research before visiting a city rich in museums and ancient architecture. Say 旅行研究, museum homework, travel research, or give a city name with cultural prep intent to start. The flow parses city and flags, runs an extensive Research pass described as twelve parallel search agents covering history, museums, architecture, and archaeology, then refines content and generates org-mode documentation plus PNG pocket cards. It explicitly chains other skills (Research, ContentAnalysis, ljg-card) in NATIVE mode. For Prism’s builder audience it is adjacent: it demonstrates a rigorous multi-agent research pipeline you might emulate for competitive or domain research, but its deliverables are travel artifacts. Intermediate complexity due to orchestration and bilingual search expectations.3.6kinstalls5Ljg Writesljg-writes is a Chinese-language writing engine skill for builders who publish critical, layered essays rather than bullet-point posts. It constrains output to Org-mode with Denote-style timestamps and tags, ASCII art only for diagrams, and a voice that mixes surgical analysis with direct, friend-to-friend tone. The process forces one clear thesis, then successive cuts—counter-question, mechanism drill-down, and inversion—so each section deepens rather than repeats. Articles aim for roughly 1000–1500 Chinese characters of substantive argument. Prism catalogs it under Grow because solo founders use long-form writing for newsletters, blogs, and community trust while shipping products. Invoke when you have a single opinion to stress-test in public and want the agent to follow a repeatable editorial ritual instead of generic chat prose.3.5kinstalls6Ljg Plainljg-plain (白) is a cognitive-atom agent skill that rewrites any source into language a smart twelve-year-old could repeat, without prescribing a single essay template. Solo builders install it when READMEs, specs, support macros, or investor updates sound like committee copy and you need human speech instead. The skill locks a floor: no unexplained jargon, one idea per sentence, concrete nouns and verbs, and an opening line that earns the next sentence. It optionally emits Org notes under Denote conventions with filetags and identifiers, which suits Emacs-centric indie workflows more than casual chat-only users. Because triggers span explain-this and plain across the journey, treat it as journey-wide clarification—validate scope docs, build internal docs, grow support answers, and operate runbooks all benefit. It does not fact-check claims; you still verify numbers and legal wording after simplification.3.5kinstalls7Ljg Learnljg-learn is a journey-wide agent skill that acts as a concept anatomist for solo builders and indie makers who need more than a Wikipedia gloss before writing specs, marketing copy, or architecture docs. Invoke it whenever you ask to explain, dissect, or deeply understand a term or idea—especially via triggers like 解剖概念, 概念解剖, explain concept, learn concept, or /ljg-learn. The workflow anchors the common definition and misconceptions, runs eight tight cuts (historical lineage, dialectical tension, phenomenological scene, etymology and semantic neighbors, formalization with failure bounds, existential impact, aesthetic image, and meta-metaphor critique), then introspects and compresses into a formula, one sentence, and ASCII structure diagram. Output is always pure org-mode with zero markdown exceptions, which suits Emacs-centric knowledge bases and long-lived research notes. It is procedural knowledge for agents, not an integration: your coding agent follows the ritual and produces a citable learning artifact you can reuse across validate, build, and launch decisions.3.5kinstalls8Ljg Investljg-invest generates a deep investment analysis report for solo builders, angels, or founder-adjacent operators who evaluate projects from conversation logs, pitch decks, or sparse descriptions. It explicitly avoids traditional ratio stacking; the core question is whether the venture creates new order or merely shuttles old order. The report uses a five-section frame that flexes in length based on evidence—starting with a dimensional table and a non-market-label赛道 definition, then a order-creation machine verdict via flywheel motion, behavior under shocks, and whether growth is pushed or pulled. Known companies can be named alone while the agent enriches via research workflows. It suits validate-phase diligence on acquisitions, partnerships, or your own pivot, and idea-phase framing when comparing opportunities. It is not a substitute for licensed financial advice or automated trading.3.4kinstalls9Ljg Rankljg-rank is a journey-wide analytical skill that decomposes any domain into its true rank—the count of independent generators that can regenerate the field’s phenomena. It rejects vague “core principles” in favor of explanations that are hard to vary: change one root and the whole predictive story fails. Before digging, it forces you to name the domain’s basic assumptions (the ceiling you are not allowed to question) versus the root generators (the floor you discover by recursion). The workflow is seven mental steps from listing phenomena through candidate generators and recursive “why” questioning to independence and falsifiability. Solo and indie builders use it when entering a new niche, auditing a competitor narrative, or choosing what their product must embody—not what blog posts say matters. Output is written rank plus stated assumptions, suitable as a foundation for scope docs, positioning, or agent research briefs.3.4kinstalls10Ljg Wordljg-word is an agent skill that turns a single English word into a deep-dive mastery brief instead of a shallow translation. Solo and indie builders shipping English-first SaaS, docs, and marketing copy invoke it when a term feels fuzzy—whether in an API README, a landing headline, or support macros—and need imagery, core semantics, and usage insight they can actually reuse. The skill formats a Markdown report with phonetics and Chinese gloss, then walks through original physical imagery, a compact core-imagery formula, and an explanatory section with bolded keywords, ending in a bilingual philosophical one-liner. It fits Prism’s multi-phase journey: Build docs for precision, Grow content for voice, and Launch SEO for semantically rich phrasing. Complexity stays beginner: pass one word; no repo setup. Pattern is generative research prose grounded in word semantics, optimized for agents that already render Markdown in chat.3.4kinstalls11Ljg Paper Flowljg-paper-flow is a workflow skill that chains ljg-paper and ljg-card for every source the user supplies—arXiv URLs, PDF paths, or paper titles. It targets solo researchers and indie builders who want agent-native batch paper literacy without manually invoking two skills per document. After extracting the paper list from chat, each paper runs in its own subagent: step A produces an org interpretation file; step B feeds that org into ljg-card with the requested visual mold. Cross-paper parallelism is allowed; within-paper steps stay strictly ordered so card content reflects the generated analysis, not raw PDF text. Quality bars and red lines from the child skills remain in force. Ideal when users say 论文流, paper flow, or ask for both reading notes and shareable sketchnote assets in one pass.3.4kinstalls12Ljg Skill Mapljg-skill-map is a lightweight meta utility for Claude Code users who accumulate dozens of skills and lose track of what is installed. The skill is implemented as a shell script that walks the standard skills directory, reads each SKILL.md frontmatter block, and prints a single JSON array suitable for piping into jq, docs generators, or Prism-style catalogs. It captures name, version, user_invocable, and a short description derived from the first sentence when descriptions run long or use YAML block scalars. For solo builders, that turns an opaque folder tree into an auditable manifest you can diff after upgrades, share with teammates, or feed into automation that warns about duplicate or stale skills. It does not invoke agents or modify files—read-only filesystem access only—making it a practical first step before refactors, security reviews, or documenting which skills apply to Validate versus Ship phases.3.3kinstalls13Ljg Relationshipljg-relationship is a Mandarin-first agent skill that acts as a relationship structure analyst for users who say 关系分析, 分析关系, relationship, or describe a specific interpersonal problem. It is not a generic chatbot coach: it refuses prescriptive fixes and instead uses dialogue to help someone see dynamics they cannot name—splitting structural issues (power, exchange, boundaries, stage, narrative) from repeating scripts (transference and unconscious patterns). The flow begins by 接住 the user’s situation, then asks whether they want tactical clarity on one incident or understanding why the same conflict keeps returning. A concrete story with sequence and dialogue is required before depth work. Language stays analogical rather than jargon-heavy, and the agent tracks resistance when topics are skipped. Solo builders benefit when founder–co-founder, boss, or family pressure threatens focus and shipping judgment, though the skill is personal-coaching oriented rather than a dev integration. Outputs are insight and visualized structure, not HR policies or legal guidance.3.3kinstalls14Ljg Word Flowljg-word-flow is an agent workflow skill that chains ljg-word and ljg-card -i so solo builders and creators get both a rich English word breakdown and a matching infograph card in one invocation. Users pass one or more space-separated words; for each word the agent runs ljg-word to produce Markdown semantics in chat, then feeds that result into ljg-card -i to render a PNG saved under ~/Downloads/. The skill explicitly runs in NATIVE mode as a skill pipeline only, without the broader Algorithm seven-step ritual. Multiple words can be processed in parallel via subagents while preserving irreversible ordering inside each word. It fits indie educators, newsletter writers, and language-focused content makers who want consistent depth-plus-visual assets instead of hand-building slides. Quality bars from the upstream skills remain unchanged; the flow skill only orchestrates handoff and reporting.3.3kinstalls15Ljg Paper Riverljg-paper-river is a research agent skill that implements a “read backward” method for academic papers: starting from one target paper, it recursively finds predecessor work it critiques or improves (up to five levels), then scans forward for newer progress, and writes a cohesive problem-evolution story. Solo builders exploring AI features, ML-heavy products, or research-backed positioning use it when a single PDF is insufficient and they need the intellectual river—why each generation tackled which bottleneck and what gap remained. Output follows a structured Org document with tags, source metadata, ASCII maps, staged narratives, frontier extension, compressed one-screen diagram, and actionable research heuristics. It complements generic web search by forcing problem-centric linkage rather than isolated summaries. Complexity is intermediate because it assumes the agent can access paper URLs, citations, and venue metadata reliably.3.2kinstalls16Ljg ThinkLjg-think (追本之箭) is a journey-wide agent skill that treats one user-supplied idea, phenomenon, or problem as a tunnel to drill until logic, physics, human structure, or paradox stops further why-questions. Solo builders invoke it when triggers like think deep, drill down, 想透, or 本质是什么 appear—not when they want a landscape survey of many options. The agent acts as a single-direction arrow: name each stratum in a few characters, expose a crack (contradiction or question) at the layer bottom, and switch explanatory frameworks downward until irreducible ground. Output is narrative free-fall, not a flat report. After running date for timestamps, it writes ~/Documents/notes/{timestamp}--追本-{theme}__think.org in org-mode without markdown. Use in Idea before positioning, in Validate before locking scope, in Build when a design choice needs root-cause clarity, or in Grow when messaging must rest on a defensible why.3kinstalls17Ljg Readljg-read is a user-invocable companion-reading agent skill for builders and knowledge workers who want to understand hard texts—not outsource them to a summary. When you say 伴读, read with me, or hand it a URL, PDF, or pasted essay, it fetches or reads the source, detects language, and runs a phased workflow. Phase 0 produces a global map: a one-sentence anchor, paragraph classes marked as skeleton, muscle, or connective tissue, and an ASCII structural overview agents label as their judgment so you can override. English sources get a faithful Chinese translation layer; Chinese sources skip straight to structure and questioning. The skill insists on org-mode typography and ASCII-only diagrams so output stays portable in notes tools indie founders already use. It is ideal in Idea when researching competitors, policy, or technical papers, and in Grow when studying distribution or lifecycle content deeply. It deliberately does not replace your thinking—it manufactures the questions you would not have asked alone.2.8kinstalls18Ljg Presentljg-present is a演讲铸造器—an outline-to-stage HTML generator for solo builders who already wrote structure in orgmode (`*`, lists, tables, emphasis) and want manifesto-style visuals without a manifesto rewrite. Unlike deck tools that compress copy to one word per slide, it maps org hierarchy to pages: top-level stars become accent cover beats, nested headings become theme pages, paragraphs split only for physical pagination, and `*bold*` / `~code~` pick up highlight colors. Pick black, red, or yellow for the entire run (or infer from file tags), or switch to cyber-hacker styling with `--cyber`. Triggers include 讲这个, present, 铸成演示, slogan, and 按 outline 美化. Output lands as a single downloadable HTML file so you can open it locally or screen-share during validation sessions, launch rehearsals, or internal readouts. The skill explicitly rejects rewriting: it is a renderer, not a copywriter—stage presence stays with the speaker.2.3kinstalls19Ljg Qaljg-qa is an agent skill that teaches how to design question chains and answers that carry real argumentative weight instead of collapsing into definitions or MBA bullet lists. Solo builders and indie operators use it when turning research, product decisions, or technical topics into FAQ pages, launch narratives, course segments, or long-form posts where each Q must force a specific kind of reasoning. The methodology mixes action, contrast, causality, and boundary questions so readers get method, cost, and limits in one flow. On the answer side, it enforces a one-sentence takeaway, a visual formalization line using simple symbols, short derivation steps, and explicit failure modes. It fits agents drafting documentation or growth content whenever generic Q&A would sound hollow or unfalsifiable.1.8kinstalls20Ljg Pushljg-push is a maintainer workflow skill for authors who keep personal ljg-* agent skills under ~/.claude/skills and need them published to the public ljg-skills repository without manual branch juggling. It enforces a two-branch contract: master stays org-mode with .org files and org headers, while md receives automated markdown conversion with YAML frontmatter and bold syntax updates. Before any push, a README consistency gate compares every local ljg-* directory against README mentions and stops the release if documentation drift would confuse installers. The skill is explicitly scoped—it does not push arbitrary repos or non-ljg skills. Solo builders who fork or mirror this pattern can treat it as a template for gated skill publishing, but out of the box it only targets lijigang’s fixed repo URL and Workflow/Push.md plus Tools/Push.sh execution path.1.7kinstalls21Ljg Bookljg-book is an agent skill for solo builders and indie makers who learn from books but refuse generic summaries. When someone shares a title or asks to 拆书 / analyze what a book is really arguing, the skill runs a disciplined two-phase ritual: first a calm structural extraction (question versus topic, unspoken assumptions, the author’s worldview frame, conclusions, and one distilled takeaway), then an interactive “reference map” so you can walk the author’s lens and predict novel situations outside the pages. Tone is direct and analytical—not praise, not plot recap. It pairs with ljg-think, ljg-paper, and ljg-rank for adjacent depth work. Use it early in the journey when market or strategy reading must become decisions, when validating scope from a domain canon, or when turning reading into growth content with a defensible mental model rather than bullet notes.830installs