Claude Code Vault

An Obsidian vault that documents how Claude Code works under the hood. Written for developers who read code, not marketing copy.

Every concept maps back to what it becomes in the API call.

What this is

Knowledge in 7 layers, from the raw API call to plugins and permissions. Each layer builds on the previous one.

The layers

Layer 1: The API call        what system + messages + tools actually contains
Layer 2: Tools               how the model requests actions through JSON schemas
Layer 3: Autonomy            the agent loop, sub-agents, custom agent files
Layer 4: Programmed behavior skills, commands, and writing instructions that work
Layer 5: Identity             CLAUDE.md as system prompt, hooks as event listeners
Layer 6: Extension           MCP servers, plugins, packaging
Layer 7: Boundaries          permissions, context window, operational limits

How to use it

As a learning path (linear)

Open 00 Claude Code Vault.md. Read the layers top to bottom. Each note has Previous/Next links at the bottom for sequential navigation.

As a reference (non-linear)

Open any note directly. Cross-references link to related concepts. The graph view shows how everything connects.

On GitHub (without Obsidian)

The notes are plain markdown. You can read them in any order on GitHub. Internal links use Obsidian [[wikilink]] syntax, which will not resolve as clickable links on GitHub, but the filenames are readable enough to find the referenced note.

Setup

1. Clone this repo
2. Open the folder as an Obsidian vault
3. Start at 00 Claude Code Vault.md

Display settings (inline title, CSS snippet, excluded files) are pre-configured in .obsidian/. No manual setup needed.

Sources

This vault draws from:

• The Claude Code system prompt (visible in any session or via claude --debug)
• Anthropic official documentation
• Anthropic prompt engineering tutorial
• Claude Code plugin ecosystem (Apache 2.0)
• Observable behavior from production use

Short fragments of the Claude Code system prompt are quoted throughout the vault for analysis and commentary. These are used under fair use for educational purposes.

Several files are reproduced in full for educational analysis (Apache 2.0 and MIT licensed).

License

The original content in this vault (explanations, analysis, structure, diagrams) is licensed under CC BY 4.0.

Third-party content is used under its original license:

Content	Source	License
code-reviewer agent	plugins/feature-dev	Apache 2.0
commit command	plugins/commit-commands	Apache 2.0
code-review command	plugins/code-review	Apache 2.0
frontend-design skill	plugins/frontend-design	Apache 2.0
playground skill	plugins/playground	Apache 2.0
claude-code-action CLAUDE.md	claude-code-action	MIT
System prompt fragments	Claude Code CLI	Fair use (short quotes with analysis)
Documentation references	docs.anthropic.com	Paraphrased, not reproduced

See NOTICE for full attribution details.

Contributing

This is a living document. If you find something wrong, outdated, or missing, open an issue or submit a PR.