Mental Model
Knowledge is for deliberate capture of reference material. Unlike memory (which extracts info from conversation), knowledge is explicit: user asks to save something.
Trigger Patterns
| User Intent |
Action |
Destination |
| "Save this link" |
Store URL |
bookmarks.jsonl |
| "Make a note about X" |
Create note |
notes/YYYY-MM-DD_topic.md |
| "Remember this code pattern" |
Save snippet |
snippets/{lang}/ |
What Goes Where
.knowledge/
├── bookmarks.jsonl # URLs, articles, docs
├── notes/
│ └── YYYY-MM-DD_topic.md # Meeting notes, ideas, summaries
└── snippets/
├── typescript/ # Language-specific patterns
├── python/
└── shell/
Entry Schemas
Bookmark:
{"id":"bm_{{timestamp}}","url":"{{URL}}","title":"{{page title}}","tags":["{{keywords}}"],"savedAt":"{{ISO8601}}"}
Note (Markdown + YAML):
---
title: {{Topic}}
date: {{YYYY-MM-DD}}
tags: [{{keywords}}]
---
# {{Topic}}
{{content}}
Snippet:
// @title: {{Pattern Name}}
// @tags: {{keyword1}}, {{keyword2}}
// @created: {{YYYY-MM-DD}}
{{code}}
Anti-Patterns (NEVER)
- Don't save transient information (today's weather, temporary URLs)
- Don't duplicate - check if already exists before saving
- Don't save without user's explicit request (use
memory skill for that)
Action Pattern
User says "保存这个链接" or "Save this":
- Extract URL, title, context
- Generate relevant tags
fs_write to appropriate destination- Confirm what was saved
