← 返回 Skills 市场
284
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install openclaw-soul-publish
功能描述
OpenClaw 自我进化框架一键部署。安装宪法(AGENTS.md)、可进化灵魂(SOUL.md)、心跳系统、PARA三层记忆架构、目标管理，并通过场景化对话引导用户定义 Agent 性格。自动配置 EvoClaw（审批制进化）和 Self-Improving Agent（自主学习）。触发场景："setup o...
安全使用建议
Before installing, do these checks: 1) Inspect scripts/preflight_check.py and the validators yourself; run them manually rather than letting the agent execute them automatically. 2) Back up any existing ~/.openclaw/workspace/ files (AGENTS.md, SOUL.md, etc.). 3) Do NOT paste API keys into the chat — if you want external feeds, set the environment variables yourself in a controlled way rather than letting the skill write to your shell profile. 4) Decide governance explicitly: set EvoClaw to supervised/advisory if you want human approval for identity changes (the docs are inconsistent about defaults). 5) Consider testing this in a throwaway workspace first (copy the package to a temp workspace) to observe behavior. 6) If you are uncomfortable with an agent having autonomous identity‑evolution rights, do not install or keep EvoClaw in 'autonomous' mode.
功能分析
Type: OpenClaw Skill
Name: openclaw-soul-publish
Version: 1.2.0
The bundle implements a highly complex 'Self-Evolution' framework that requires extensive privileges, including shell execution, modification of the agent's core constitution (AGENTS.md), and the installation of external skills. A significant security risk is present in 'fallback/evoclaw/tools/soul-viz.py', which starts a local web server with a POST endpoint ('/save-soul') that allows overwriting the agent's identity file (SOUL.md) without authentication or CSRF protection. While these capabilities are aligned with the stated purpose of creating an evolving AI, the combination of a local write-enabled web server and autonomous modification of the agent's primary instruction set creates a substantial attack surface.
能力评估
Purpose & Capability
The skill's stated purpose (deploy a self‑evolution framework and optional integrations) matches the actions in SKILL.md: creating workspace files, configuring EvoClaw and Self‑Improving components, and optionally integrating social feeds. However, some capabilities are more powerful than a simple 'template deployer' — it includes full fallback skill bundles, validators, and CLI-style config flows that ask the agent to manage external API keys and heartbeats. Those behaviors are plausible for the stated goal but are heavier than a minimal deploy tool.
Instruction Scope
Instructions direct the agent to read and write many user files (workspace templates, evoclaw-state.json, memory directories, and files under ~/self-improving/), copy fallback skill directories into the user's skills folder, and run bundled Python scripts (scripts/preflight_check.py and validators). The fallback configure flow explicitly tells the agent to accept raw API keys and automatically write EXPORT lines into the user's shell profile — this is sensitive and surprising behavior for an automated skill and expands the agent's runtime authority beyond just copying templates.
Install Mechanism
No external download/install spec is present (low risk from network fetches). However, the package includes local executable artifacts (preflight_check.py, validators, tools) which the instructions tell the agent to run. The fallback approach copies local directories into the user's workspace — copying bundled code that may then be executed is riskier than pure instruction-only skills.
Credentials
Registry metadata lists no required env vars, but the runtime docs instruct the agent to request and persist external service API keys (MOLTBOOK_API_KEY, X_BEARER_TOKEN) and to write them into shell profile files automatically. Requesting API keys for optional external feeds is understandable, but automatically storing raw keys into ~/.bashrc/.zshrc and exporting them without explicit, informed consent is disproportionate and risky. The skill also reads/writes ~/.openclaw/openclaw.json and other user files — expected for deployment but worth highlighting.
Persistence & Privilege
The skill is not force‑installed (always:false) and is user-invocable, which is normal. However, the framework it installs (EvoClaw/Self‑Improving) is explicitly designed for ongoing autonomous evolution and heartbeat-driven updates; some docs and templates reference 'autonomous' defaults while other deployment steps write governance as 'advisory' — these inconsistencies affect how much autonomy the agent will get post‑install. Because autonomous model invocation is possible, you should verify governance settings before enabling external feeds or allowing automated proposal application.
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 openclaw-soul-publish - 安装完成后，直接呼叫该 Skill 的名称或使用
/openclaw-soul-publish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
v1.2.0
Three-level fallback system for zero-dependency deployment. Inline Self-Improving and Identity Evolution protocols in AGENTS.md. Full offline skill files in fallback/ directory.
元数据
常见问题
Openclaw Soul Publish 是什么？
OpenClaw 自我进化框架一键部署。安装宪法(AGENTS.md)、可进化灵魂(SOUL.md)、心跳系统、PARA三层记忆架构、目标管理，并通过场景化对话引导用户定义 Agent 性格。自动配置 EvoClaw（审批制进化）和 Self-Improving Agent（自主学习）。触发场景："setup o... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 284 次。
如何安装 Openclaw Soul Publish？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install openclaw-soul-publish」即可一键安装，无需额外配置。
Openclaw Soul Publish 是免费的吗？
是的，Openclaw Soul Publish 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Openclaw Soul Publish 支持哪些平台？
Openclaw Soul Publish 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（linux, darwin, win32）。
谁开发了 Openclaw Soul Publish？
由 Kianzzz（@kianzzz）开发并维护，当前版本 v1.2.0。
推荐 Skills