什么是 知识管理技能？

知识管理技能提供了一个强大的框架，用于管理 AI 代理生成的长期记忆和见解。通过解析 MEMORY.md 和每日日志等源文件，它会自动将信息归类为研究、教训和决策等不同类型。

该工具将扁平的文本文件转换为高度组织化的目录结构，确保关键的项目知识始终可访问且可操作。对于使用 Openclaw 技能的开发人员来说，这是一个必不可少的组件，他们需要维护一个干净、幂等且可进行版本控制的文档系统，而无需人工干预。

下载入口:https://github.com/openclaw/skills/tree/main/skills/claireaicodes/knowledge-management

安装与下载

1. ClawHub CLI

从源直接安装技能的最快方式。

2. 手动安装

将技能文件夹复制到以下位置之一

优先级：工作区 > 本地 > 内置

3. 提示词安装

将此提示词复制到 OpenClaw 即可自动安装。

知识管理技能 应用场景

• 自动化从原始每日日志到永久、分类的本地知识库的转换。
• 为特定内容类型（如项目决策或技术教程）生成结构化索引。
• 将本地 AI 记忆与版本控制的文档系统同步，以便团队协作。
• 清理孤立或冗余的记忆文件，以维持简洁准确的知识库。

1. 该技能会扫描配置的工作区以查找源文件，包括主要的 MEMORY.md 和每日记忆日志。
2. 每个条目都会被解析并传递给分类引擎，由其确定内容类型、领域、确定性和影响。
3. 为每个条目生成唯一的 8 字符内容哈希，以方便去重并防止文件名冲突。
4. 经过验证的条目将作为单独的 Markdown 文件写入输出目录中特定类型的文件夹（如 /Research 或 /Lesson）。
5. 内部同步状态将被更新以跟踪已处理的条目，从而在未来的运行中实现快速、幂等的同步。

知识管理技能 配置指南

配置您的工作区环境变量或使用默认路径开始使用 Openclaw 技能工具：

export OPENCLAWORKSPACE=~/.openclaw/workspace

可选：预先创建目录结构以准备环境：

mkdir -p ~/.openclaw/workspace/memory/KM/{Research,Decision,Insight,Lesson,Pattern,Project,Reference,Tutorial}

运行您的第一次同步以处理过去一周的条目：

km sync --days_back 7

知识管理技能 数据架构与分类体系

该技能根据内容分类将数据组织成层级文件夹结构。每个生成的文件都包含一个 YAML 前置数据块，以便与其他 Openclaw 技能保持元数据兼容性。

命名规范： YYYYMMDDTHHMM_标题_哈希.md

name: knowledge-management
description: Organize and classify OpenClaw knowledge entries into local folders by content type (Research, Decision, Insight, Lesson, Pattern, Project, Reference, Tutorial).
homepage: https://github.com/ClaireAICodes/openclaw-skill-knowledge-management
metadata: { "openclaw": { "emoji": "??", "requires": { "bins": ["km"] } } }

Knowledge Management Skill (Local Storage)

Organize your OpenClaw memory files into a structured local knowledge base. Automatically parses MEMORY.md and daily memory files, classifies entries by content type, and stores each as a timestamped markdown file in the appropriate folder.

Available Tools

Core Commands

• km sync [options] - Sync memory entries to local files
• km classify [options] - Parse and classify without storing (JSON output)
• km summarize [options] - Generate index files for each content type
• km cleanup [options] - Remove orphaned files
• km list_types - List all available content types

Setup

No API keys needed! The skill uses two locations:

• Input Workspace: Where MEMORY.md and memory/ daily files are read from.
• Output Directory: Where organized folders (Research/, Decision/, etc.) are written.

Both are detected automatically:

Input Workspace (source files)

1. OPENCLAWORKSPACE environment variable
2. --workspace CLI argument
3. Current working directory (if it contains MEMORY.md)
4. Default: ~/.openclaw/workspace

Output Directory (organized files)

1. --output-dir CLI argument (relative to workspace or absolute)
2. Default: /memory/KM

The skill will create the output directory and all content-type folders automatically.

If you want to pre-create:

mkdir -p ~/.openclaw/workspace/memory/KM/{Research,Decision,Insight,Lesson,Pattern,Project,Reference,Tutorial}

Usage Examples

Default locations (input at workspace root, output in memory/KM)

# From any directory (workspace auto-detected)
km sync --days_back 7 --cleanup

Custom input workspace and output directory

km sync --workspace /custom/input/workspace --output-dir /custom/output/KM --days_back 7

Using environment variables

export OPENCLAWORKSPACE=/custom/input/workspace
km sync --output-dir /custom/output/KM --days_back 7

Dry run (preview only)

km sync --dry_run --days_back 1

Classify entries and export JSON

km classify --days_back 3 > entries.json

Generate index files (default: output directory)

km summarize
# or specify different location
km summarize --output_dir ~/some/other/folder

Preview orphan cleanup

km cleanup --dry_run

List content types

km list_types

Storage Structure

Assuming default configuration:

• Input workspace: ~/.openclaw/workspace
• Output directory: ~/.openclaw/workspace/memory/KM

~/.openclaw/workspace/
├── MEMORY.md                (source file - you edit this)
├── memory/                  (daily memory files)
│   ├── 2025-02-11.md
│   ├── 2025-02-12.md
│   └── ...
└── memory/KM/               (organized output by the skill)
    ├── local-sync-state.json
    ├── local-sync-log.md
    ├── Research/
    │   ├── 20260215T1448_Title_Here_HASH.md
    │   └── ...
    ├── Decision/
    ├── Insight/
    ├── Lesson/
    ├── Pattern/
    ├── Project/
    ├── Reference/
    ├── Tutorial/
    ├── Research_Index.md
    ├── Decision_Index.md
    └── ... (other index files)

File Naming

Format: YYYYMMDDTHHMM_Title_With_Underscores_8CHARHASH.md

The 8-character content hash suffix prevents filename collisions when titles are identical but content differs.

File Content (YAML Frontmatter)

---
title: "Protocol Name"
content_type: "Research"
domain: "OpenClaw"
certainty: "Verified"
impact: "Medium"
confidence_score: 8
tags: ["AI", "Automation"]
source: "MEMORY.md"
source_file: "MEMORY.md"
date: "2026-02-11"
content_hash: "e4b30e75d0f5a662"
---

Entry body content starts here...

How It Works

1. Parses MEMORY.md and recent daily memory/*.md files
2. Classifies each entry (content type, domain, certainty, impact, tags, confidence)
3. Computes content hash for deduplication
4. Checks sync state (memory/local-sync-state.json) to skip already synced entries
5. Writes to appropriate folder with timestamp + hash filename
6. Updates state mapping (hash → filepath)
7. Optional cleanup removes files not in state

Classification Logic

• Content Type: Keyword matching (Research, Lesson, Decision, Pattern, Tutorial, Reference, Insight)
• Domain: Contextual inference (AI Models, OpenClaw, Cost, Trading, etc.)
• Certainty: Based on language (Verified, Likely, Speculative, Opinion)
• Impact: Importance indicators (High, Medium, Low, Negligible)
• Tags: Auto-extracted from predefined keyword map
• Confidence Score: 1–10 heuristic (source credibility, length, data mentions)

Customize by editing the EntryClassifier class in index-local.js.

State Management

memory/local-sync-state.json maps content hashes to file paths:

{
  "e4b30e75d0f5a662": "/path/to/Research/202602151440_Title_e4b30e75.md"
}

This enables idempotent syncs and fast duplicate detection.

Do not edit manually unless recovering from corruption.

Cron Integration

Automate daily syncs:

openclaw cron add r
  --name "Daily Knowledge Sync" r
  --cron "0 5 * * *" r
  --tz "Asia/Singapore" r
  --session isolated r
  --message "km sync --days_back 7"

Note: By default, the skill reads MEMORY.md from ~/.openclaw/workspace and writes organized files to ~/.openclaw/workspace/memory/KM. Use --workspace or --output-dir to customize these locations.

Troubleshooting

"km: command not found"

• Run npm link in the skill directory, or add ~/workspace/bin to PATH.

No entries found

• Ensure MEMORY.md uses ## section headers and ### entry titles within recognized sections.

Files not created

• Check write permissions; run with --verbose.

Old entries not syncing

• They may already be in state. Clear memory/KM/local-sync-state.json to force re-sync (caution: may duplicate files).

Duplicate files

• Run km cleanup to remove orphans, then km sync to create missing files.

Version: 2.0.0 Changed: 2026-02-15 — Switched from Notion to local storage, added hash suffixes for uniqueness. Author: Claire (OpenClaw Agent) License: MIT