← 返回 Skills 市场
learning-engine
作者
mupengi-bot
· GitHub ↗
· v1.0.1
3487
总下载
0
收藏
34
当前安装
2
版本数
在 OpenClaw 中安装
/install learning-engine
功能描述
Auto-analyze mistake and success patterns and reflect in skills
安全使用建议
Install only if you intentionally want an agent to learn from local memory files and potentially change other skills. Before enabling it, require manual approval with visible diffs for every skill-file update, disable automatic hooks unless needed, redact sensitive log content, keep backups, and set retention limits for generated learning artifacts.
功能分析
Type: OpenClaw Skill
Name: learning-engine
Version: 1.0.1
The `SKILL.md` file describes a 'learning-engine' skill designed to auto-update other skill definitions by injecting 'learned rules' into `skills/{skill-name}/SKILL.md` files. These rules are derived from various sources, including error logs. This capability creates a significant prompt injection vulnerability and potential Remote Code Execution (RCE) risk, as attacker-controlled input (e.g., crafted error messages) could be processed, converted into a 'rule', and then injected into other skill definitions, leading to unauthorized execution by the agent. While the skill's stated purpose is self-improvement, the mechanism for modifying other skills without explicit sanitization of learned content is a critical design flaw that allows for malicious exploitation.
能力评估
Purpose & Capability
Learning from errors, evaluations, and performance data fits the stated purpose, and the artifact discloses memory reports, learned rules, events, and skill updates. The concern is not hidden code, but that learned natural-language content can become future agent instructions.
Instruction Scope
The skill directs automatic updates to `skills/{skill-name}/SKILL.md` based on logs and evaluations, but does not require explicit user approval, diff review, target allowlists, sanitization, provenance checks, or rollback before changing other skills.
Install Mechanism
The package contains only `SKILL.md` and no executable scripts; static scan is clean. However, the instructions describe hook and scheduled integration, so safe use depends on external configuration and user-controlled enablement.
Credentials
Reading memory/error, self-eval, and performance stores is proportionate for a learning skill, but writing consolidated reports and propagating rules into multiple skill files is broad and can preserve sensitive or manipulated content.
Persistence & Privilege
The skill creates persistent learned-rule files, weekly and meta reports, event files, and modified skill instructions, with scheduled and hook-triggered operation described but no retention limits, stop controls, or approval gates.
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 learning-engine - 安装完成后，直接呼叫该 Skill 的名称或使用
/learning-engine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
v1.0.1
Docs update
v1.0.0
learning-engine v1.0.0
- Initial release automates learning from mistakes and successes to generate actionable rules.
- Extracts patterns from error logs, self-evaluations, and performance data.
- Auto-generates and injects learned rules into relevant skill SKILL.md files.
- Publishes weekly learning reports and "lesson learned" events.
- Integrates with hook-engine for error, self-eval, performance, and scheduled triggers.
- Includes meta learning and roadmap for future enhancements.
元数据
常见问题
learning-engine 是什么？
Auto-analyze mistake and success patterns and reflect in skills. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 3487 次。
如何安装 learning-engine？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install learning-engine」即可一键安装，无需额外配置。
learning-engine 是免费的吗？
是的，learning-engine 完全免费（开源免费），可自由下载、安装和使用。
learning-engine 支持哪些平台？
learning-engine 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 learning-engine？
由 mupengi-bot（@mupengi-bot）开发并维护，当前版本 v1.0.1。
推荐 Skills