← 返回 Skills 市场
306
总下载
0
收藏
2
当前安装
2
版本数
在 OpenClaw 中安装
/install moltcare-open
功能描述
Install and configure the MoltCare Agent Framework - a four-layer configuration system (SOUL/AGENTS/USER/MEMORY) with three-layer trigger architecture (Exact...
安全使用建议
This package is largely what it says: it installs templates that OpenClaw will auto-load each session and it adds a weekly cron trigger. Before installing: 1) Review all assets (AGENTS.md, SOUL.md, etc.) to confirm you accept the behavioral rules (especially 'act first, ask later' / PUA escalation). 2) Back up your existing ~/.openclaw/workspace — the installer copies files into that folder root and could overwrite things. 3) Inspect scripts/install.sh and avoid running any remote curl | bash commands from the README; prefer running the included, local install.sh after review. 4) If you do not want persistent behavior, do not install or remove the added files and crontab entry after testing (crontab -l | grep '检查token优化' to find/remove). 5) If you want tighter control, install in a sandbox account or test environment first and confirm the agent's actions match your expectations. If you want, I can (a) produce a diff of what install.sh will change given your HOME value, or (b) highlight any lines in the assets that you should edit to reduce autonomous behavior.
功能分析
Type: OpenClaw Skill
Name: moltcare-open
Version: 3.2.0
The MoltCare-Open skill bundle is a comprehensive configuration framework for OpenClaw agents, focusing on task persistence, multi-expert decision-making, and token cost optimization. While it introduces high-autonomy behaviors (e.g., 'Act first, ask later' and 'PUA' pressure escalation levels in SOUL.md and AGENTS.md), it includes explicit safety guardrails that prohibit the agent from accessing credentials, modifying permissions, or executing destructive commands like 'rm -rf'. The installation script (install.sh) sets up a standard workspace and a cron job for weekly token audits, but contains no hidden payloads, obfuscation, or data exfiltration logic.
能力评估
Purpose & Capability
The name and description match what the files and install.sh do: copy template docs into ~/.openclaw/workspace, create a memory dir, and optionally configure a weekly cron trigger. There are no unrelated env vars, binaries, or mysterious remote credential requests.
Instruction Scope
SKILL.md and assets instruct the agent to auto-load files from the workspace root, read and write memory files, run token-audit triggers, and use OpenClaw tools (read/edit/exec/memory_search). These actions are consistent with a configuration framework. However the content also prescribes behavioral policies (PUA 'act first, ask later', pressure escalation, automatic persistent recording rules) that change agent behavior beyond mere template installation and could cause the agent to take more autonomous actions than some users expect.
Install Mechanism
There is no remote installer executed by the provided scripts: scripts/install.sh only copies bundled assets into the user's home and sets a cron job. That's low-risk. BUT the README suggests an alternate install via curl https://raw.githubusercontent.com/... | bash which is a high-risk pattern; if users follow that, they would execute remote code arbitrarily. The included install.sh itself is local and readable.
Credentials
The skill does not request environment variables, credentials, or unusual config paths. It writes to ~/.openclaw/workspace and ~/.openclaw/workspace/memory and adds a crontab entry — these are proportional for a workspace/template installer. Note: the assets reference locations of credential files in documentation (e.g., ~/.config/openclaw/credentials, ~/.env) as reminders; the skill does not directly access them, but placing files in the auto-loaded workspace can cause the agent to read user-provided content you put there.
Persistence & Privilege
The installer is persistent: it copies multiple policy/behavior documents into ~/.openclaw/workspace which OpenClaw will auto-load each session, and it installs a weekly cron entry that writes to a workspace trigger file. That persistent, auto-loaded content will change agent behavior on every session (including the PUA problem-solving rules). Combined with normal autonomous invocation, this increases the potential for the agent to perform actions the user didn't anticipate. The skill does not alter other skills or system-wide settings beyond the user's workspace and crontab.
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 moltcare-open - 安装完成后，直接呼叫该 Skill 的名称或使用
/moltcare-open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
v3.2.0
**MoltCare-Open v3.2.0 – Major update with token optimization and new audit features**
- Added layered task execution and daily token optimization audit, reducing OpenClaw’s routine token usage.
- Introduced new config/assets files: BEST_PRACTICES.md, CONFIG_CHECKLIST.md, IDENTITY.md, TOKEN_AUDIT.md, TOOLS.md, and token-audit-template.md.
- Updated AGENTS.md and documentation to include cost-efficient L0–L3 task layering and automatic audit scheduling.
- Installation and file-loading instructions clarified: templates must now be in the workspace root for auto-loading; memory/audit templates are loaded on demand.
- Weekly token audit is now auto-configured (cron job); manual trigger supported.
- Revised SKILL.md and install script to reflect new templates, audit workflow, and best practices guidance.
v3.1.0
Initial release: Three-layer trigger architecture with PUA problem-solving framework
元数据
常见问题
Skill 是什么？
Install and configure the MoltCare Agent Framework - a four-layer configuration system (SOUL/AGENTS/USER/MEMORY) with three-layer trigger architecture (Exact... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 306 次。
如何安装 Skill？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install moltcare-open」即可一键安装，无需额外配置。
Skill 是免费的吗？
是的，Skill 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Skill 支持哪些平台？
Skill 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 Skill？
由 useens（@useens）开发并维护，当前版本 v3.2.0。
推荐 Skills