Content
-Native AI Wiki
[Public Wiki](https://pennypny163.github.io/agent-native-llm-wiki/) ·
[MCP Documentation](meta/mcp.md)
This is an AI knowledge management system for humans and agents. Word documents are upstream materials, and `sources/imported/` contains local, reproducible source layers, while other directories hold curated, verified, and linked knowledge.
## Getting Started
-M Topic Map](maps/llm.md)
- [LLM Engineer Learning Path](learning-paths/llm-engineer.md)
- [RAG Topic Map](maps/rag.md)
- [Agent Topic Map](maps/agent.md)
- [Agent Engineer Learning Path](learning-paths/agent-engineer.md)
AI Product Management Topic Map](maps-management.md)
- [AI Interview Prep](learning-paths/ai-interview-prep.md)
Knowledge Base Maintenance](meta/)
- [Content Model: Di + Evergreen](meta/content)
- [Evergreen Core Judgments](evergreen/)
- [Publishing Wiki](meta/publishing.md)
- [MCP Knowledge Service/mcp.md)
## Content Hierarchy
| Level | Directory | Purpose | |
|---|---|---|---|
| Raw Materials | Root directory `.docx` | Raw context | No |
| Source Layer | `sources/imported/` | Structured, searchable copies of DOCX | No, regenerable |
| Evidence Layer | `evidence/` | Source lists, verification records, conflict records | Yes |
| Canonical Knowledge Layer | `concepts/`, `explanations/`, `reference/` | De-duplicated best current expressions | Yes |
| Action Layer | `how-to/`, `learning-paths/` | Paths for tasks and learning | Yes |
| Navigation Layer | `maps/` | Thematic relationships and entry points## Common Commands
```bash
# Rebuild source layer, index, and verify
make build_wiki
# Incremental import and list affected knowledge pages
make apply_updates
# Full verification
make verify_wiki
#
make search Q="KV Cache"
# Generate public Wiki for publishing docs-prepare
# Local preview after installing requirements-docs.txt
makeserve
# Test and demo MCP knowledge service
make mcp-test
cp-demo
```
## Core Principles
1. Do not modify raw.
2. Keep only one `canonical` page per topic; link other pages to it.
3. Conclusions must link back to source documents and sections.
4. Tutorials, guides, explanations, and references are organized separately.
5. Volatile information like models, frameworks, prices, and APIs must be labeled with verification dates.
Git tracks public Markdown, scripts, and rules; raw Word documents, private source text, source images, and local indexes are reproducible materials, not in the regular Git history.
Code, automation, and MCP implementation use the [MIT License](LICENSE-CODE); knowledge content and upstream material rights boundaries are in
[CONTENT-NOTICE](CONTENT-NOTICE.md).
MCP Store