← 返回 Skills 市场
Persistent Memory
作者
Jakebot-ops
· GitHub ↗
· v3.0.0
1210
总下载
0
收藏
7
当前安装
3
版本数
在 OpenClaw 中安装
/install persistent-memory
功能描述
Three-layer persistent memory system (Markdown + ChromaDB vectors + NetworkX knowledge graph) for long-term agent recall across sessions. One-command setup w...
安全使用建议
What to do before running this skill:
- Inspect the files scripts/configure_openclaw.py and the inline configure code in unified_setup.sh to confirm the exact changes (which extraPaths will be added).
- Run configure_openclaw.py with --dry-run first (or run the standalone script instead of unified_setup.sh) to see proposed edits and ensure you have a good backup of ~/.openclaw/openclaw.json.
- If you must use unified_setup.sh, run it in a non-production workspace and avoid running as a privileged user; review and commit a backup of your OpenClaw config beforehand.
- Be aware the setup will pip-install packages and the sentence-transformers model will download weights from the network; if you have air-gapped or constrained environments, perform installs manually.
- Consider removing or excluding sensitive files (IDENTITY.md, USER.md, etc.) from the configured extraPaths if you do not want them indexed by the agent.
- If you prefer manual control, run the setup steps one-by-one (create venv, copy scripts, run indexer, then run configure_openclaw.py with --backup) rather than the one-command unified installer.
功能分析
Type: OpenClaw Skill
Name: persistent-memory
Version: 3.0.0
The skill is classified as suspicious due to its capability to modify the core OpenClaw application configuration (`~/.openclaw/openclaw.json`) and execute `openclaw` CLI commands (e.g., `openclaw gateway restart`) via `scripts/unified_setup.sh` and `scripts/configure_openclaw.py`. While the stated purpose is benign (to improve agent compliance by making OpenClaw index critical workspace files like SOUL.md and AGENTS.md), these actions involve significant system modification and shell execution. This presents a high-privilege capability that, if exploited or misused, could lead to vulnerabilities, even though there is no clear evidence of intentional malicious behavior (e.g., data exfiltration, backdoors, or malicious prompt injection) within the provided code.
能力评估
Purpose & Capability
The code and SKILL.md match the described purpose: parsing workspace markdown, building ChromaDB vectors and a NetworkX graph, and integrating those memories into OpenClaw. The indexer, search, graph, and auto_retrieve scripts implement the three-layer memory, and the configuration script targets OpenClaw memorySearch as promised.
Instruction Scope
The runtime instructions tell the user to run a one-command installer (unified_setup.sh) that not only creates a local venv and installs dependencies but also locates and modifies OpenClaw configuration files and attempts to restart OpenClaw. The inline Python used by unified_setup.sh updates ~/.openclaw/openclaw.json or ./openclaw.json directly (without invoking the configure_openclaw.py backup path), which is a system-level change beyond simply adding local memory files. This is in-scope for integrating with OpenClaw but is high-impact and performed without an explicit interactive confirmation or guaranteed backup in the unified path.
Install Mechanism
No registry install spec was provided; setup is performed by the provided shell scripts which create a Python venv and run pip to install pinned packages (sentence-transformers, chromadb, networkx). This is a moderate-risk, expected mechanism for a Python-based tool. Note: running the code will download model weights (sentence-transformers) from external servers at runtime.
Credentials
The skill requests no environment variables or external credentials. It does, however, add many workspace files (SOUL.md, AGENTS.md, IDENTITY.md, USER.md, reference/) to OpenClaw's memorySearch extraPaths — this will expose workspace-local directives and potentially sensitive local files to the memory index. That behavior is coherent with the stated goal but should be reviewed because it broadens what the agent will automatically read/index.
Persistence & Privilege
The skill modifies OpenClaw's configuration (a system-wide agent setting) and attempts to restart OpenClaw via the 'openclaw' CLI. The installer path bundled in unified_setup.sh makes these modifications non-interactively (no backup code path is invoked in that inline configure routine). While the action aligns with the skill's function, changing another tool's global config is high-privilege and should be done with explicit user consent and backups.
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 persistent-memory - 安装完成后，直接呼叫该 Skill 的名称或使用
/persistent-memory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
v3.0.0
🚀 ONE-COMMAND SETUP: Eliminates installation friction with unified_setup.sh that automatically handles 3-layer memory system creation, Python dependencies, OpenClaw memorySearch integration, and initial indexing. No more manual configuration steps.
v2.0.0
🚨 CRITICAL UPDATE: Fixes directive compliance vulnerability. Adds OpenClaw memorySearch configuration to prevent agents ignoring workspace directives (SOUL.md, AGENTS.md, etc.). All v1.x users must upgrade immediately. Includes configure_openclaw.py script for automatic integration.
v1.0.0
Initial release: three-layer persistent memory (Markdown + ChromaDB vectors + NetworkX knowledge graph). One-command setup, semantic search, sync watchdog.
元数据
常见问题
Persistent Memory 是什么？
Three-layer persistent memory system (Markdown + ChromaDB vectors + NetworkX knowledge graph) for long-term agent recall across sessions. One-command setup w... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 1210 次。
如何安装 Persistent Memory？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install persistent-memory」即可一键安装，无需额外配置。
Persistent Memory 是免费的吗？
是的，Persistent Memory 完全免费（开源免费），可自由下载、安装和使用。
Persistent Memory 支持哪些平台？
Persistent Memory 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 Persistent Memory？
由 Jakebot-ops（@jakebot-ops）开发并维护，当前版本 v3.0.0。
推荐 Skills