AgentSkillsCN

build-knowledge

使用 Markdown 文件构建并管理持久化知识库（第二大脑）。当需要存储、检索或链接复杂概念、项目背景，或在不同会话间“记住”相关信息时，可使用此技能。触发条件包括：(1) “记住这一点”；(2) “构建知识库”；(3) “记录这一概念”；(4) 将研究发现归纳为长期记忆。

SKILL.md
--- frontmatter
name: build-knowledge
description: "Build and manage a persistent knowledge base (second brain) using Markdown files. Use when you need to store, retrieve, or link complex concepts, project context, or 'remember' information across sessions. Triggers on: (1) 'Remember this', (2) 'Build a knowledge base', (3) 'Document this concept', (4) Summarizing findings into long-term memory."

Build Knowledge

You are the curator of a persistent knowledge base stored in .agent_memory. Your goal is to create a densely linked "second brain" that avoids information silos.

Core Principles

  • •The Golden Rule of Linking: NEVER create orphaned files. Every new note MUST be linked from at least one other note (usually 000_Index.md or a related topic).
  • •Search Before Write: Always search for existing topics before creating a new node to prevent duplication.
  • •Wikilinks: Use [[Note_Name]] for internal links.

Librarian Tool (Scripts)

Use the Librarian script via uv to manage the knowledge base.

1. Search for existing knowledge

Finds files containing specific keywords using case-insensitive grep and synonym expansion.

bash
uv run python .agent/skills/build-knowledge/scripts/librarian.py search --keywords "keyword1" "keyword2"

2. Read a note

Retrieves the full content of a specific note.

bash
uv run python .agent/skills/build-knowledge/scripts/librarian.py read --filename "Note_Name.md"

3. Create a new note

Stores new information. Fails if the file already exists.

bash
uv run python .agent/skills/build-knowledge/scripts/librarian.py create --filename "Note_Name.md" --content "# Title\nContent here..."

4. Append to a note

Adds details to an existing topic.

bash
uv run python .agent/skills/build-knowledge/scripts/librarian.py append --filename "Note_Name.md" --content "Additional info..."

Dual-Write Protocol (Graph Sync)

To ensure the Agent's "active memory" matches the "archived memory", you MUST synchronize key concepts to the MCP Graph Memory if the tools are available.

Triggers:

  • •Creating a Note -> Create a corresponding Entity.
  • •Linking Notes -> Create a corresponding Relation.

Procedure:

  1. •Check if mcp_memory_create_entities is available.
  2. •Files First: Execute the librarian.py file operations as usual.
  3. •Graph Second:
    • •Create an Entity for the main topic (Name = Note Title, Type = Concept/Topic).
    • •Add an Observation summarizing the content (approx. 1 sentence).
    • •Create Relations linking it to other entities (e.g., NewTopic relates_to Index).

Standard Workflow

  1. •Analyze: User says "Remember X".
  2. •Search: search --keywords "X" to check for existing context.
  3. •Branching Logic:
    • •If exists: read it, then append new info. Update Graph with new Observation.
    • •If new: create the node. Create Entity in Graph.
  4. •Link: read the parent/index note and append a link [[X]] to ensure discovery. Create Relation in Graph.
  5. •Confirm: Tell the user where the knowledge is stored (File + Graph).

Troubleshooting & Setup

If using for the first time or if the script fails due to missing dependencies, see references/setup.md.