Zettelkasten Knowledge Gardener

Zettelkasten Knowledge Gardener

Theory Base

This skill combines five compatible ideas:

• Zettelkasten: keep notes atomic, linked, and composable instead of storing large undifferentiated documents
• How to Take Smart Notes: move from capture -> literature notes -> durable notes
• Evergreen notes: write notes as living ideas that can be revised and reused
• PARA: organize the garden by action context: Projects, Areas, Resources, Archives
• Progressive Summarization: keep useful compression layers so future retrieval is fast

Use the theory as a practical system, not as ideology. The goal is not to imitate a perfect note-taking religion. The goal is to help the user build a knowledge garden that stays useful over time.

Purpose

Turn messy material such as:

• conversations
• reading notes
• research fragments
• project lessons
• personal wiki pages
• saved links and highlights

into reusable knowledge assets such as:

• atomic notes
• evergreen notes
• MOC or index pages
• context packs for future prompting
• review and pruning queues

This skill is for knowledge maintenance, not just summarization. It should always leave behind artifacts that can be reused later.

When to Use

Use this skill whenever the user wants to:

• organize or clean up a knowledge base
• turn a book, article set, or conversation log into durable notes
• create a second-brain style system
• refresh a personal wiki or notes vault
• extract reusable context from a project or research stream
• merge duplicate notes or identify stale knowledge
• prepare a prompt-ready context pack from existing notes

Strong trigger phrases include:

• "organize my knowledge base"
• "整理知识库"
• "整理读书笔记"
• "build a second brain"
• "update my personal wiki"
• "turn this into evergreen notes"
• "create an MOC"
• "make this reusable later"

Core Principles

1. Prefer durable knowledge over raw storage

Do not simply archive what the user gives you. Distill what is worth keeping.

2. Keep notes atomic

Each note should hold one idea, one claim, one principle, one case, or one open question. Avoid packing many unrelated ideas into one note.

3. Separate evidence from interpretation

Every durable artifact should distinguish:

• Facts: what the source directly supports
• Inferences: what is concluded or generalized
• Open Questions: what is unresolved
• Review Trigger: what should cause the note to be revisited

4. Titles should help retrieval

Prefer note titles that are specific and statement-like. A good evergreen title can often be read as a claim.

Bad:

• "RAG notes"
• "Book highlights"

Better:

• "Retrieval quality degrades when chunk boundaries ignore task intent"
• "Prompt context packs should preserve source confidence, not just conclusions"

5. Link before filing

Do not create isolated notes if they clearly relate to existing ideas. Surface the most meaningful links and clusters.

6. Prune aggressively

Mark notes as stale, duplicate, contradictory, or low-value when the evidence supports that judgment.

Operating Modes

Choose the lightest useful mode.

Mode A: Quick Harvest

Use for one article, one conversation, or a small note batch.

Output:

• atomic notes
• 1 small index section
• short review queue

Mode B: Reading Distillation

Use when the user finished a book, paper set, or long article cluster.

Output:

• literature notes
• evergreen notes
• concept links
• reading-derived context pack

Mode C: Garden Refresh

Use when the user already has a note system and wants cleanup or restructuring.

Output:

• duplicate map
• stale note list
• conflict list
• PARA placement recommendations
• pruning and review plan

Mode D: Prompt Context Pack

Use when the user wants to turn existing knowledge into something an agent can reuse immediately.

Output:

• compact background brief
• key principles
• known uncertainties
• source confidence notes
• suggested prompts or future retrieval hooks

Knowledge Unit Model

When you create or reshape notes, use this internal model:

## [Note Title]
Type: Fact / Principle / Case / Open Question
Source: [conversation, book, article, project, or "not specified"]

Facts
- ...

Inferences
- ...

Open Questions
- ...

Links
- [[Related note]]
- [[Related MOC]]

Review Trigger
- Revisit when ...

Not every user-facing output needs to show every field in full, but the skill should reason with these distinctions.

Workflow

Step 1: Clarify the gardening goal

Identify:

• what material the user is giving you
• whether the goal is capture, distillation, cleanup, or reuse
• whether the user wants a note system, a context pack, or both

Step 2: Separate signal from note clutter

Break the material into units such as:

• facts
• principles
• examples
• decisions
• open questions
• repeated but low-value fragments

Do not give equal weight to everything.

Step 3: Create atomic notes

Rewrite high-value material into reusable units. Each unit should be understandable without the entire original source sitting beside it.

Step 4: Link and cluster

Show how the notes connect:

• parent concept -> child idea
• general principle -> concrete example
• question -> possible answer
• conflicting interpretations

When useful, create a small MOC or index page instead of only isolated notes.

Step 5: Place into PARA

Recommend where the outputs belong:

• Projects: active, time-bound efforts
• Areas: ongoing responsibilities
• Resources: reference material
• Archives: inactive but worth preserving

Do not force PARA labels if the input is too sparse; say so clearly.

Step 6: Add compression layers

Use progressive summarization:

• raw source or source pointer
• distilled bullets
• evergreen note
• prompt-ready context pack

This lets the same knowledge be useful at different levels of time pressure.

Step 7: Leave maintenance hooks

Always end with at least one of:

• a pruning list
• a review queue
• conflict checks
• stale note warnings
• future retrieval hooks

Output Format

Use this structure unless the user asks for something more specific.

# Knowledge Garden Update

## Goal
- [What this gardening pass tried to accomplish]

## Inputs
- [Sources or material processed]

## Atomic Notes
- [Note title]: [1-2 sentence summary]

## Evergreen Notes
- [Claim-like note title]: [Why it matters]

## Links or MOC
- [Cluster, index page, or concept relationship]

## PARA Placement
- Projects:
- Areas:
- Resources:
- Archives:

## Context Pack
- Facts:
- Inferences:
- Open Questions:
- Review Triggers:

## Pruning or Review Queue
- [Duplicate, stale, contradictory, or follow-up item]

If the user wants a lighter output, compress the same logic rather than dropping the distinctions.

Handling Messy Inputs

If the source material is weak or incomplete:

• say what is directly supported
• mark what is inferred
• do not invent links that are not justified
• explicitly note when a note is not yet evergreen quality

Useful phrases:

• "Worth capturing, but still too raw for an evergreen note."
• "Duplicate idea with lower signal than the existing note."
• "Potential link, but not well supported yet."
• "Archive unless this becomes relevant to an active project."

Quality Bar

A strong response from this skill should:

• leave behind reusable artifacts instead of a generic summary
• preserve the line between facts and inference
• create notes that can be linked, updated, or pruned later
• recommend structure without becoming bureaucratic
• help the user's knowledge get easier to retrieve over time