An AI-augmented personal knowledge management system built on Obsidian and Claude Code. It uses the PARA method for organization, a layered Wiki for knowledge weaving, and AI-driven contextual retrieval — so your notes become a working memory that grows with you.
You collect notes, bookmarks, highlights, and ideas — but they sit in folders and never connect. When you need them, you can't find them. When you find them, they lack context.
This template solves that by giving you:
- A clear structure — PARA separates actionable work from reference material
- A knowledge layer — Wiki weaves raw materials into entities, concepts, and analyses
- An AI layer — Claude Code understands your knowledge graph and retrieves relevant context automatically
Think of it as a memory palace where AI is your librarian.
Second Brain/
├── Inbox/ # Capture everything here first
├── Projects/ # Active work with deadlines
├── Areas/ # Ongoing responsibilities (no end date)
├── Resources/ # Raw materials — articles, reports, notes
├── Archives/ # Completed or paused items
├── Identity/ # Who you are — mission, context, preferences
│ ├── TELOS.md # Your purpose and goals
│ ├── CONTEXT.md # Current focus and active projects
│ └── PROFILE.md # Communication style and preferences
├── System/ # Operational support
│ └── working-memory/ # Tasks, daily logs, operating rules
│ ├── tasks.md
│ ├── OPERATING_RULES.md
│ └── daily/ # Daily interaction logs
├── Wiki/ # Woven knowledge (the real value)
│ ├── index.md # Knowledge map — entry point for AI retrieval
│ ├── entities/ # People, companies, projects (concrete things)
│ ├── concepts/ # Ideas, frameworks, methodologies (abstract things)
│ ├── analyses/ # Deep-dive research reports
│ └── overviews/ # Domain landscape surveys
└── .claude/
└── CLAUDE.md # AI system prompt — the brain's operating system
Save anything to Inbox/ or Resources/. Articles, meeting notes, random thoughts — don't organize yet, just capture.
AI processes raw materials and weaves them into the Wiki:
Raw article about EigenLayer (Resources/)
↓ AI processes
Wiki/entities/EigenLayer.md — structured entity page
Wiki/concepts/Restaking.md — concept extraction
Wiki/analyses/EigenLayer-Thesis — investment analysis
↓ AI connects
[[EigenLayer]] ←→ [[Ethereum]] ←→ [[DeFi-Core-Concepts]]
Each Wiki page includes:
- Frontmatter with source tracking, tags, and creation date
- Bidirectional links connecting related knowledge
- Source attribution back to raw materials in Resources/
When you ask a question, AI follows a retrieval path:
You: "What's the investment thesis for EigenLayer?"
↓
AI reads Wiki/index.md → finds [[entities/EigenLayer]]
↓
AI reads the entity page → follows links to analyses, concepts
↓
AI responds using YOUR knowledge, not generic training data
The system maintains itself:
- Frontmatter consistency — metadata stays in sync with file locations
- Index updates — Wiki/index.md reflects current knowledge
- Knowledge smells — AI detects orphaned notes, stale content, broken links
- Identity tracking — preferences and operating rules accumulate over time
| Tool | Purpose | Install |
|---|---|---|
| Obsidian | Local-first knowledge vault | Download from official site |
| Claude Code | AI layer — powers retrieval, weaving, and maintenance | npm install -g @anthropic-ai/claude-code |
| Plugin | Purpose | Install Command |
|---|---|---|
| obsidian-skills | Obsidian Flavored Markdown, JSON Canvas, and Bases format support | /plugin marketplace add kepano/obsidian-skills && /plugin install obsidian@obsidian-skills |
| Plugin | Purpose | Install Command |
|---|---|---|
| superpowers | Enhanced planning, debugging, and task management skills | /plugin install superpowers@claude-plugins-official |
This template ships with four custom skills in System/skills/ that power the core workflows:
| Skill | Trigger | What It Does |
|---|---|---|
| intake | /intake or "intake" |
Scans Inbox → archives to Resources/ → weaves knowledge into Wiki/ |
| maintain | /maintain or "maintain" |
Full vault health scan: structure, content, weaving progress |
| digest | /digest or "digest" |
Reviews AI-observed preference candidates for user approval |
| writing | /writing or "writing" |
Loads your writing style rules, then writes in your voice |
These skills are plain Markdown files — edit them to match your workflow.
| Plugin | Purpose |
|---|---|
| Dataview | Query and display note metadata |
| Templater | Advanced templates for new notes |
| QuickAdd | Fast capture workflows |
| Graph Analysis | Visualize knowledge connections |
git clone https://github.com/Nowhitestar/second-brain-template.git "My Second Brain"- Open Obsidian → "Open folder as vault"
- Select the cloned directory
- Enable community plugins (for Dataview, Templater, etc.)
# Navigate to your vault
cd "My Second Brain"
# Required: Obsidian format support
/plugin marketplace add kepano/obsidian-skills
/plugin install obsidian@obsidian-skills
# Recommended: Enhanced planning & debugging
/plugin install superpowers@claude-plugins-officialThe .claude/CLAUDE.md file is pre-configured. Customize it:
- Edit
Identity/TELOS.md— define your mission and goals - Edit
Identity/CONTEXT.md— describe your current focus - Edit
Identity/PROFILE.md— set your communication preferences
# Open Claude Code in your vault directory
claude
# Try these:
# "Help me organize my Inbox"
# "What do I know about [topic]?"
# "Summarize my active projects"The .claude/CLAUDE.md file turns Claude Code into a knowledge-aware assistant. Key capabilities:
AI automatically loads relevant knowledge based on what you're discussing:
检索路径 (Retrieval Path):
1. Match topic → PARA directories
2. Read directory README → find member files
3. Match by title and summary
4. Cross-check Resources/ for cross-domain content
5. Search Wiki/index.md → load matching entities/concepts
| Layer | What | When | Where |
|---|---|---|---|
| Daily fragments | Conversation logs | Auto, per session | System/working-memory/daily/ |
| Weekly distillation | Pattern recognition | Weekly trigger | Candidate pool |
| Stable preferences | Confirmed habits | User approval | Identity/PROFILE.md |
Inbox/Resources → AI processes → Wiki/ (entities, concepts, analyses)
↓
index.md updated
↓
Bidirectional links created
↓
Frontmatter validated
| Type | Action | Approval |
|---|---|---|
| A | File archiving, tag completion, README sync | Auto |
| B | Identity changes, new categories | Required |
| C | Top-level directory changes, system prompt changes | Required |
Edit the README.md in each PARA directory to match your domains. The template provides a starting point — rename, add, or remove directories as needed.
Edit the <RETRIEVAL> section in .claude/CLAUDE.md to change search priorities. For example, if you work primarily in one domain, prioritize that directory.
The default Wiki has four categories: entities, concepts, analyses, overviews. Add more if your work requires it (e.g., decisions/, people/, tools/).
L1 .claude/CLAUDE.md — Global rules (this is L2/L3 folded up)
L2 Directory README.md — Local map, member list
L3 Note frontmatter — Content contract, ownership
Each layer is self-similar: L1 is L2 compressed, L2 is L3 compressed. Directory READMEs serve as distributed indexes — no separate index file needed (except Wiki/index.md for AI retrieval).
Structure and content must stay in sync. Any change in one must be reflected in the other.
This means:
- Add a file → update the directory README
- Move a file → update frontmatter + old/new READMEs
- Delete a file → update the directory README
- Create a directory → create its README
| Smell | Description |
|---|---|
| Island | Valuable content with no links to other knowledge |
| Redundancy | Same idea duplicated across multiple files |
| Staleness | Outdated content still marked as active |
| Chaos | Large backlog piled up in Inbox |
| Breakage | Links pointing to deleted/moved files |
| Emptiness | Directory exists but has no README or members |
- Local-first: All data stays in your Obsidian vault
- No cloud dependency: Works offline, syncs via iCloud/Git
- Identity protection: Changes to Identity files require explicit approval
- Change audit trail: B/C type changes go through approval queue
Inspired by:
- Tiago Forte's PARA Method
- MemPalace — the memory palace concept for AI
- Obsidian — local-first knowledge management
- Claude Code — AI-powered development
MIT License — See LICENSE for details.