Hands-On: Build Your Library from Zero in 30 min

No coding. Follow along, check the "acceptance" after each step, and move on only when it passes.

Step 0: Prep (3 min)

• Install WorkBuddy and Obsidian (download from official sites; Obsidian is free)
• Pick 3 materials you use most (one industry article, one work doc, one study note is great)
• Decide on a name for your knowledge base (e.g., "Career Library," "My Knowledge Base")

Step 1: Create the "Library Home" (5 min)

Where to put it:

• On a non-removable target drive / workspace (dedicated data drive or workspace directory)
• Not on the system drive or desktop — system drive affects stability, desktop gets accidentally moved
• Don't move the folder after creating it (AI remembers the path; moving breaks it)

Have WorkBuddy create the folder (choose your OS; replace 【your library name】):

Windows:

In D:\, create a folder named "【your library name】."
If a folder with the same name already exists, do not overwrite or delete any of its contents; just return its current state.
Do not create any subdirectories or files in this folder this time.
Return the absolute path, verify it exists, then stop.

macOS (replace your username):

In /Users/your-username/工作区, create a folder named "【your library name】."
If a folder with the same name already exists, do not overwrite or delete any of its contents; just return its current state.
Do not create any subdirectories or files in this folder this time.
Return the absolute path, verify it exists, then stop.

Set as working directory: back in WorkBuddy, click "Select Folder" next to the input box, and set the newly created folder as the current task's working directory — all subsequent operations happen here.

Acceptance: the absolute path returned by WorkBuddy exists; File Explorer can open it; the folder is empty.

Step 2: Send the First Prompt — AI Builds Your Library (10 min)

Send the first prompt to WorkBuddy (with the working directory set to your library folder).

Watch AI's three steps:

1. Design first: based on your request, explain how the library should be designed to be reliable
2. Recommend methodology: recommend authoritative methodologies (e.g., Karpathy LLM Wiki, PARA/CODE, Zettelkasten, Evergreen Notes), summarized in plain language
3. Deep interview: if needs aren't clear enough, ask one question at a time until clear

Tips for answering the interview: the more specific your answers, the better the library fits you.

Don't know how to express your needs? Send this:

I'm not really sure what kind of knowledge base I want to build. I only know the general direction is __, and the goal is __. What questions should I think through first? Ask me one at a time to help me clarify.

After the interview, AI auto-builds the library. You'll see roughly this structure:

your-knowledge-base/
├── 库规.md (Schema)     # Discipline file, constrains AI as a "disciplined maintainer"
├── raw/                 # Source layer (immutable, AI reads only) → you drop materials here
├── wiki/                # AI-compiled knowledge layer (you read)
│   ├── index.md        # Master index + tag table (search entry)
│   ├── log.md          # Operation log
│   └── ...             # Topic knowledge, source summaries, asset cards
├── outputs/             # Retrieval/analysis outputs
└── 模板/ (Templates)    # Sedimented templates + asset card templates

Acceptance: AI designs first, recommends methodology, interviews you (one Q at a time), then auto-creates raw/ + wiki/ + Schema.md structure = success.

Step 3: Drop Materials + "Digest" (7 min)

5 ways to feed materials (pick any):

1. Click + next to input → upload file/paste screenshot/reference doc
2. Click folder icon at bottom-right of input to let WorkBuddy read from a directory
3. Paste material text directly into the input box
4. Use File Explorer to drag files into the raw/ folder
5. Send the file/folder path directly to WorkBuddy to read

Then say:

Digest

Watch AI perform this chain:

• Sources go into raw/ (original materials archived)
• Distilled into wiki/ cards (source summaries + topic knowledge)
• Index updated (index.md)
• Log recorded (prevents duplicate compiles)

Acceptance: raw/ has originals, wiki/ has cards, index.md is updated = success.

Step 4: Query with Sources — "Retrieve" from Library (5 min)

Use the query prompt, e.g.:

Query the current knowledge base: among my saved materials, what reusable methods are there for 【time management / productivity】?
Distinguish confirmed, inferred, and pending, and provide clickable source links.

Acceptance: AI pulls cards from your library to answer (not from its own knowledge), each with sources that link back to the original.

Step 5: View Your Library in Obsidian (5 min)

WorkBuddy shows you conversations; now switch to Obsidian to see "what it actually did."

Open the Same Library

Open Obsidian → "Open folder as vault" → select the same folder (WorkBuddy's working directory). Obsidian and WorkBuddy view the same files.

View Structure

The left file tree shows raw/ (originals), wiki/ (knowledge cards), Schema.md (rules) — this is the three-layer structure.

Three-Stage Trace Chain

Core: any conclusion can be traced back to the original.

1. Open wiki/index.md (master index) → click a card link
2. In the card, click the source summary link
3. In the summary, click the original material link → back to raw/ original

Mantra: for any important conclusion, ask three times: who said it, where, and what's the original text.

Index → card → source summary → original — all four stages must be clickable end-to-end for the knowledge base to be trustworthy.

View Graph (Auxiliary)

Open any card in wiki/ → click the graph icon in the left sidebar (two dots connected by a line), or press Ctrl+P (Cmd+P on Mac) and search "graph" in the command palette → see connections between cards.

The graph is an auxiliary tool for checking links, not the knowledge base itself; a lack of fancy visuals doesn't affect usage.

Acceptance: you can click from index.md all the way back to the original (four stages connected) = success.

Minimal Loop Complete

At this point, your minimal loop is done: build → store → retrieve. Congratulations — you have a library that grows on its own.