← Back to Skills Marketplace
101
Downloads
0
Stars
0
Active Installs
2
Versions
Install in OpenClaw
/install memory-extraction
Description
自动识别对话中的实体、关系和事实观察，实时更新知识图谱以维护用户记忆信息。
Usage Guidance
This skill looks like a memory manager but has gaps and privacy implications you should resolve before using it. Ask the publisher for the missing code (knowledge_graph_manager.py) and a homepage/source repo so you can review how data is stored and protected. Confirm exactly where memory files are written and who can read them. Consider these safeguards before installing: (1) require explicit user consent before recording PII, (2) restrict which fields are stored (avoid emails/relations if not needed), (3) put the memory file in a controlled directory with access controls and encryption, (4) avoid automatic System Prompt injection unless you review and approve the prompt changes, and (5) only enable the skill in trusted contexts. If you can’t review implementation details or control what is recorded, treat the skill as high-risk and do not enable it for sensitive conversations.
Capability Analysis
Type: OpenClaw Skill
Name: memory-extraction
Version: 1.0.1
The skill bundle implements a knowledge graph-based memory system designed to extract and store entities, relations, and observations from conversations. While it instructs the agent to collect personal information (identity, preferences, and goals), this behavior is transparently documented as the skill's primary purpose and utilizes local storage (memory/knowledge-graph.jsonl). There are no indicators of data exfiltration, malicious execution, or harmful prompt injection in SKILL.md or _meta.json.
Capability Assessment
Purpose & Capability
The SKILL.md describes extracting entities/relations/observations and writing a knowledge graph, which fits the name. However the instructions reference specific local files and functions (scripts/knowledge_graph_manager.py, memory/knowledge-graph.jsonl, APIs like search_nodes/read_graph/create_entities) that are not included in the package and were not declared as required config paths. That mismatch (instructions expecting local code/storage while the skill is instruction-only) is incoherent and could lead the agent to try to read/write unexpected local files.
Instruction Scope
The runtime instructions direct the agent to proactively identify and persist wide-ranging personal data (identity: age/gender/location/education, behaviors, preferences, goals, and relationships up to 3 degrees). They also instruct modifying the system prompt and to perform reads/writes to local paths. Proactive capture and storage of sensitive PII without clear consent controls is a privacy risk and expands scope beyond a narrow helper.
Install Mechanism
No install spec and no code files are included — lowest installation risk. There is nothing being downloaded or executed by an installer in the package metadata.
Credentials
The skill requests no environment variables or credentials, which is proportionate. However, despite no credential requests, the instructions require persistent local storage of potentially sensitive personal information (emails, locations, relationships). Lack of declared config paths for those storage locations is discrepant and increases risk of unintended local-file access.
Persistence & Privilege
always is false (good). But the SKILL.md instructs integration into the agent's System Prompt and to persist memories to local files, which grants ongoing behavioral influence and persistent data storage. The skill does not request system-wide privileges explicitly, but the combination of system-prompt modification and persistent local storage is a privilege-sensitivity concern and should be handled with explicit user consent and access controls.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install memory-extraction - After installation, invoke the skill by name or use
/memory-extraction - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.1
- 更新示例实体和关系名称为更具泛化性的表达（如“用户”、“张三”、“北京”）。
- 优化示例邮箱、时区格式，更贴近实际用途和规范。
- 文档内容细微调整，提升规范性和易读性。
- 不涉及功能代码改动，仅文档梳理和示例优化。
v1.0.0
memory-extraction 1.0.0
- 自动从对话中提取实体(Entity)、关系(Relation)与观察(Observation)并写入知识图谱
- 支持用户、项目、技能、工具、偏好等多种实体类型的自动识别与创建
- 关系识别包括 owns、uses、prefers、located_at、named、created_on、deployed_to 等
- 观测值要求原子化、一事实一条，便于验证与检索
- 提供手动 API 操作与自动对话内规则
- 集成 System Prompt，规范 Agent 记忆管理行为
Metadata
Frequently Asked Questions
What is Memory Extraction?
自动识别对话中的实体、关系和事实观察，实时更新知识图谱以维护用户记忆信息。 It is an AI Agent Skill for Claude Code / OpenClaw, with 101 downloads so far.
How do I install Memory Extraction?
Run "/install memory-extraction" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Memory Extraction free?
Yes, Memory Extraction is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Memory Extraction support?
Memory Extraction is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Memory Extraction?
It is built and maintained by paibwhgs (@paibwhgs); the current version is v1.0.1.
More Skills