← 返回 Skills 市场

EvoMind - Agent记忆与进化引擎

Name: EvoMind - Agent记忆与进化引擎
Author: thebuddha5566

作者 thebuddha5566 · GitHub ↗ · v1.0.2 · MIT-0

cross-platform ✓ 安全检测通过

总下载

当前安装

版本数

在 OpenClaw 中安装

/install evomind

功能描述

Agent进化操作系统 — 五层记忆×自我进化引擎。可直接安装运行，非纯文档。

使用说明 (SKILL.md)

EvoMind — Agent记忆与进化引擎

不是文档，是代码。 pip install 后直接 from evomind import MemoryCore。

快速开始

pip install -r requirements.txt  # 零外部依赖，仅标准库

# CLI 演示
python scripts/memory_core.py health
python scripts/memory_core.py remember "用户偏好深色模式"
python scripts/memory_core.py recall "深色"

# Python API
python -c "
from scripts.memory_core import MemoryCore
mem = MemoryCore()
mem.remember('Python version: 3.11', priority=5, tags=['env'])
mem.skill_save('web-scraper', open('SKILL.md').read(), description='网页抓取')
mem.curate()
print(mem.stats())
"

五层架构

层	功能	方法
L1 持久记忆	关键事实自动存档，≤20KB	`remember()`, `forget()`, `recall_l1()`
L2 技能层	成功方案永久固化	`skill_save()`, `skill_load()`, `skill_find()`
L3 缓存层	会话临时数据，TTL过期	`cache_set()`, `cache_get()`, `cache_clear()`
L4 策展层	每日自动审查优化	`curate()`, `curator_log()`
L5 索引层	FTS5全文搜索，跨层召回	`recall()`, `stats()`

核心特性

✅ 跨会话持久记忆 — SQLite存储，关了再开还在
✅ 自动归档 — L1超20KB自动归档最早条目到l1_archive
✅ 冗余检测 — L4 curator自动发现重复技能（Jaccard≥0.6）
✅ 自我进化 — skill_save() + skill_load() 形成改进循环
✅ 零依赖 — 仅Python标准库+内置sqlite3
✅ FTS5全文搜索 — 跨L1/L2/L5统一语义检索

生产验证

来自 Hermes Agent 生产环境实战验证（2026年6月起），非理论设计。已在真实Agent系统中运行五层全架构。

许可证

MIT — 免费使用、修改、分发。

安全使用建议

Install this only if you want the agent to keep local memory across sessions. Avoid storing secrets, regulated data, or sensitive personal details unless you are comfortable with them being retained in the SQLite database and archive table; review or delete ~/.agent_memory.db when you want to clear stored memory.

能力评估

✓ Purpose & Capability

The artifacts consistently describe an agent memory engine with durable facts, saved skills, cache, curation, and search; the Python code implements those local storage features without unrelated capabilities.

ℹ Instruction Scope

Runtime actions are user-invoked through Python APIs or CLI commands. The skill advertises persistent memory and automatic archiving, though it could more clearly document retention implications and the default database path.

✓ Install Mechanism

requirements.txt declares zero external dependencies and the inspected code uses Python standard library modules only; no install hooks or package-manager side effects were found.

✓ Credentials

The skill writes a local SQLite database by default at ~/.agent_memory.db and can read files only when the user explicitly runs skill-save with a file path. This is proportionate for a memory tool.

ℹ Persistence & Privilege

Persistent local storage and archive tables are central to the feature set and disclosed, but users should understand that archived entries remain on disk unless manually removed.

如何使用

确保已安装 OpenClaw（本地或 Docker 部署）
在对话框中输入安装命令：/install evomind
安装完成后，直接呼叫该 Skill 的名称或使用 /evomind 触发
根据 Skill 的参数说明提供必要输入，即可获得结构化输出

版本历史

v1.0.2

修复Python 3.9兼容性（日期解析/FTS5 CJK分词/线程安全/WAL模式）。全量自测30/30通过。

v1.0.1

添加可执行代码实现：memory_core.py + cli.py。五层全通，零依赖。

v1.0.0

Initial release. 5-layer memory x self-evolution engine. First Chinese-native agent evolution system. Battle-tested from Hermes Agent production.

元数据

Slug evomind

版本 1.0.2

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 3

常见问题