Unified Memory Architect

Unified Memory Architect manages up to 1760 dream memories with layered indexing and hybrid search for fast, efficient storage and retrieval on OpenClaw Agent.

Security Scan

The skill’s stated purpose (local memory storage and query via Node scripts) broadly matches its files, but the bundle also contains agent-level guidance that tells an agent to read/write workspace memory files, commit/push changes, and act without asking — behaviors that expand its scope and require careful review before installing.

Purpose & Capability

SKILL.md, examples, and docs consistently describe a Node.js-based memory manager (query.cjs, import/migrate scripts, hybrid search). That aligns with the name and description. Minor inconsistencies: example paths vary (memory/scripts vs memory/.dreams/scripts) and the package references a GitHub repo even though 'Source' is marked unknown — these are plausibly sloppy but worth confirming.

Instruction Scope

Beyond the SKILL.md CLI/API usage, the repository contains AGENTS.md and other agent-facing docs that instruct an agent to read SOUL.md, USER.md, memory/YYYY-MM-DD.md, and MEMORY.md, to update MEMORY.md, to 'commit and push your own changes', and even to 'Don't ask permission. Just do it.' These instructions broaden the runtime scope to reading/writing many workspace files and performing network operations (git push, external chat setup). That is more powerful than a read-only query helper and could lead to unintended disclosure or external publication if scripts do network I/O.

Install Mechanism

No install spec is provided (instruction-only plus example scripts). That is lower risk from an installer perspective — nothing is downloaded or auto-executed by the platform installer. Risk surface is the code that will run if a user executes scripts locally or the agent executes them.

Credentials

The skill declares no required env vars or credentials, yet files reference workspace paths (including a symlink to /root/.openclaw/workspace/memory/.dreams per QUALITY_REPORT/REPO_PREVIEW) and expect the ability to commit/push and integrate with external services. The absence of declared credentials is not a guarantee: the scripts may rely on existing git/network credentials or the agent’s network access. This mismatch (no creds declared vs expected network/file-system access) is a proportionality concern.

Persistence & Privilege

always:false (good) but the repo explicitly encourages writing and persisting memory files, editing HEARTBEAT.md, updating MEMORY.md, and committing/pushing changes. Combined with autonomous model invocation being allowed (platform default), this increases blast radius: the skill is designed to persist data to disk and perform outbound actions if the agent follows AGENTS.md guidance.

Scan Findings in Context

[pre-scan-injection-signals-none] expected: No injection or suspicious regex findings were reported. Absence of scanner hits doesn't mean the code is safe — much of the repo is documentation and runtime behavior depends on the actual scripts (memory/scripts/*.cjs) which should be inspected for network and exec calls.

What to consider before installing

This package appears to implement a local memory manager, but the repository also contains agent-level guidance that tells an agent to read and update many workspace files and to commit/push changes — actions that can expose sensitive data or publish it remotely. Before installing or enabling this skill: 1) Inspect the actual runtime scripts (memory/scripts/*.cjs and any .cjs referenced in examples) for network calls (fetch/axios/http, sockets), child_process / exec usage, and any code that reads arbitrary filesystem paths. 2) Verify where the skill will read/write (confirm the symlink /root/.openclaw/workspace or other configured memory directory). 3) Remove or sandbox git/CI credentials (or run in an isolated environment) if you do not want accidental pushes. 4) Prefer installing only from a known upstream (confirm the GitHub repo and maintainer contact) and check commit history. 5) If you want, share the memory/scripts/*.cjs files and any verify-system/import scripts and I can inspect them specifically for network exfiltration or dangerous exec calls.

Unified Memory Architect Skill

梦境记忆系统重构与发布技能 - OpenClaw Agent Platform

技能概述

Unified Memory Architect 是一个高效的梦境记忆管理系统，为 OpenClaw Agent 提供强大的记忆存储、检索和关联能力。

版本: 1.0.0
作者: OpenClaw Team
平台: OpenClaw Agent Platform >= 2.7.0

核心能力

1. 记忆存储

• 1760个梦境记忆 - 海量记忆存储
• 7层目录结构 - 清晰的数据组织
• 60%存储节省 - 高效的数据压缩

2. 智能索引

• 49个语义标签 - 自动标签提取
• 181个实体识别 - 实体自动识别
• 多层索引 - 按类型/日期/标签/情感/语言/实体

3. 高速检索

• 5-10倍加速 - 混合搜索技术
• BM25 + 向量 + RRF - 多算法融合
• 毫秒级响应 - 优化的查询性能

使用方式

CLI 命令

# 查看统计
node memory/scripts/query.cjs stats

# 按标签查询
node memory/scripts/query.cjs tag reflection 5

# 按日期查询
node memory/scripts/query.cjs date 2026-04-12 3

# 全文搜索
node memory/scripts/query.cjs search "关键词" 3

# 按情感查询
node memory/scripts/query.cjs sentiment neutral 2

编程接口

const memory = require('./memory/scripts/query.cjs');

// 获取统计
const stats = memory.getStats();

// 按标签查询
const byTag = memory.queryByTag('reflection', 10);

// 按日期查询
const byDate = memory.queryByDate('2026-04-12', 10);

// 全文搜索
const search = memory.searchMemories('关键词', 10);

// 按情感查询
const bySentiment = memory.queryBySentiment('neutral', 10);

性能指标

指标	数值
总记忆数	1760
唯一标签	49
唯一实体	181
检索加速	5-10x
存储节省	60%

系统要求

• Node.js >= 14.0.0
• 磁盘空间 >= 100MB
• 内存 >= 512MB

相关文档

• 用户指南
• API文档
• 架构设计
• 快速开始
• 常见问题
• 故障排除

发布信息

• GitHub: https://github.com/openclaw/unified-memory-architect
• License: MIT
• Changelog: CHANGELOG