liuziyu77/anythingatlas

AnythingAtlas

Map the best way into any topic

规划任何主题的最佳学习路径

English · 简体中文

🧭 What is AnythingAtlas?

AnythingAtlas is an Agent Skill built for the first step into unfamiliar territory. You can use it with Claude Code, Codex, and other Agent Skills-compatible frameworks.

When you begin learning a new field or confronting an unfamiliar subject, the hardest part is often not finding information. It is working out what truly matters, which resources are worth your time, and where to begin.

Whether you want to enter quantitative finance, research AI agents, understand a historical event, or master a practical skill, AnythingAtlas turns scattered books, courses, papers, experts, archives, repositories, and online noise into a clear map of the field: its foundations, the resources worth learning from, how to use them, and a step-by-step plan that takes you from beginner to advanced.

🚀 How to use

All you need to do is tell AnythingAtlas what you want to learn and how much time you have. It will guide you through a focused set of questions, search broadly for carefully selected learning resources, and build a personalized study plan around your needs.

🖼️ Example Gallery

Three fields, three learning goals, and three visual themes. Select a cover to open the complete self-contained HTML atlas.

🚀 Quick start

AnythingAtlas follows the open Agent Skills format. The same SKILL.md works with Codex, Claude Code, and other Agent Skills-compatible clients; the core workflow does not need to be rewritten for each platform.

The repository is named AnythingAtlas; anything-atlas is the skill identifier in SKILL.md and the recommended installation-directory name.

Install for Codex:

mkdir -p ~/.agents/skills
git clone https://github.com/Liuziyu77/AnythingAtlas.git ~/.agents/skills/anything-atlas

Install for Claude Code:

mkdir -p ~/.claude/skills
git clone https://github.com/Liuziyu77/AnythingAtlas.git ~/.claude/skills/anything-atlas

If you have already cloned AnythingAtlas, copy or symlink that checkout into the relevant directory instead. Then invoke it explicitly, or describe your goal naturally and let the agent match the skill from its description:

Codex: $anything-atlas
Claude Code: /anything-atlas

I want to understand modern AI agents well enough to propose a research
project. I know basic Python and language models, can study eight hours
per week for twelve weeks, and prefer papers, code, and explanations in
Chinese.

If important information is missing, AnythingAtlas asks a compact set of follow-up questions before researching.

Build both deliverables from the included example:

python3 scripts/build_atlas.py \
  --input examples/sample-atlas.json \
  --output-dir /tmp/anything-atlas-output \
  --theme workshop

Available themes are atlas, scholar, archive, signal, and workshop. The CLI --theme overrides meta.theme in the JSON model; atlas is used when neither is set.

Run the structural regression suite:

python3 -m unittest discover -s tests -v

📦 Output contract

Every completed run creates:

1. anything-atlas-.md — a structured, portable, editable atlas.
2. anything-atlas-.html — a thoughtfully designed standalone presentation.

Both files use the same compact five-section sequence: confirmed brief, core guide, curated resources with user-fit tracks, detailed roadmap, and a final channel-grouped source directory. Resources assigned in a track or roadmap are clickable in both Markdown and HTML.

🎨 HTML themes

AnythingAtlas selects a visual theme from the topic, or the user can choose one. All five themes share the same content and semantic structure and work on mobile, in print, and offline.

Generated HTML uses no image logo. It credits AnythingAtlas once, as text, in the footer.

🗺️ How it works

Initial request: the topic and the time available
   ↓
Focused clarification and confirmed brief: guided questions to understand the learner and their needs
   ↓
Compact core guide: answer the real question, map concrete options, and give the first action
   ↓
Topic classification, discovery, verification, ranking, and curation
   ↓
Clickable user-fit resource tracks and a detailed personalized roadmap
   ↓
Channel-grouped source directory: official sites, books, YouTube, Bilibili, repositories, and other relevant channels
   ↓
Output: Markdown file + polished, self-contained HTML file

AnythingAtlas changes its resource selection, information channels, and verification priorities to match the type of topic:

🗂️ Repository structure

AnythingAtlas/
├── SKILL.md                         Core agent workflow
├── agents/openai.yaml               OpenAI/Codex UI and dependency metadata
├── references/                      On-demand research and output policies
├── scripts/                         Markdown/HTML rendering and validation
├── assets/
│   ├── gallery/                     README case cover artwork
│   ├── logo/logo.png                README project logo
│   └── html-template/
│       ├── atlas.html               Standalone semantic template
│       ├── atlas.css                Shared foundation for all themes
│       └── themes/                  Five visual themes and print styles
├── examples/
│   ├── sample-atlas.json            Buildable canonical example
│   ├── anything-atlas-*.md          Generated Markdown example
│   └── anything-atlas-*.html        Generated standalone HTML example
├── tests/                           Structural tests and qualitative cases
├── Design.md                        Bilingual product design
├── README.md                        English documentation
├── README.zh-CN.md                  Simplified Chinese documentation
└── LICENSE                          Apache-2.0

📚 Documentation

• Skill instructions
• Product design
• Clarification policy
• Specificity and resource-fit policy
• Topic taxonomy
• Source and channel policies
• Credibility criteria
• Roadmap schema
• Output schema
• HTML design guidelines
• Qualitative regression cases

🚧 Status

AnythingAtlas is an early functional prototype. The core skill workflow, topic-aware policies, canonical content model, dual renderers, five standalone HTML themes, shared print styles, and parity validator are implemented. Resource research still depends on the tools available to the agent running the skill.

If you have ideas for feature improvements or a better user experience, please open an issue or PR. We will work on improvements within 24 hours. If you find AnythingAtlas useful, please consider giving the project a Star—thank you for your support.

⚖️ License

Licensed under the Apache License 2.0.