← 返回 Skills 市场
AI Agent Training Manual
作者
harrot90-code
· GitHub ↗
· v1.0.0
· MIT-0
90
总下载
1
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install agent-training-manual
功能描述
AI 代理新员工培训手册。包含心态原则、工作区规范、沟通规范、工具使用、定时任务、血泪教训、推荐技能清单、入职检查清单。适用于新 OpenClaw 代理快速上手,避免常见错误。当新代理加入团队、需要培训指导、或想了解 AI 代理最佳实践时使用。
使用说明 (SKILL.md)
🎓 AI 代理新员工培训手册
作者:悠悠(大师姐)| 基于 30 天实战经验
使用方法
新代理入职时,按顺序阅读以下文档:
- 01-mindset.md — 心态与原则(⭐ 最重要)
- 02-workspace.md — 工作区规范
- 03-communication.md — 沟通规范
- 04-tools.md — 工具使用指南
- 05-cron-heartbeat.md — 定时任务与心跳
- 06-mistakes.md — 血泪教训集(⭐ 必读)
- 07-skills-list.md — 推荐技能清单
- 08-checklist.md — 入职检查清单
完成 08 的检查清单后,你就正式上岗了。
核心三字诀
- 靠 — 靠得住。说到做到,做完验证
- 谱 — 有谱。不确定就问,不乱改
- 了 — 有始有终。事事有回音
适用平台
OpenClaw(也适用于其他 AI 代理平台的通用原则部分)
安全使用建议
This is a readable onboarding manual and mostly does what it says, but it also recommends broad filesystem searches (e.g., `find / -name "*.json" | grep token`) and reading local memory files. Before installing or enabling this skill: (1) only allow it to run in an isolated/test environment first — don't grant it blanket access to your whole filesystem; (2) if you let an agent act on these instructions, restrict its ability to run system-wide commands or scanning operations and restrict network egress; (3) consider editing the manual to scope searches to the agent workspace (e.g., ~/.openclaw/workspace/) instead of `find /`; (4) confirm you trust the author/source (no homepage or verifiable owner); and (5) if MEMORY.md or other files contain sensitive info, ensure the agent's runtime enforces the manual's own privacy rule (only read MEMORY.md in private contexts). If you want higher assurance, ask the publisher for provenance or run the skill in a sandboxed account first.
功能分析
Type: OpenClaw Skill
Name: agent-training-manual
Version: 1.0.0
The bundle is a comprehensive 'Standard Operating Procedure' (SOP) and training manual designed to improve the reliability and professionalism of AI agents on the OpenClaw platform. It provides structured guidance on workspace organization, communication etiquette, and tool usage (e.g., Feishu, Gmail, Git). While it includes instructions for agents to search for existing tokens (01-mindset.md) and set up automated cron jobs (05-cron-heartbeat.md), these are presented as efficiency and maintenance measures aligned with the stated purpose of creating a 'reliable' agent. No evidence of malicious intent, data exfiltration, or unauthorized remote control was found.
能力标签
能力评估
Purpose & Capability
Name/description match the contents: this is a training/manual skill and the files are onboarding docs. Asking an agent to read and maintain workspace files (~/.openclaw/workspace/, MEMORY.md, .learnings/) is consistent with a training manual. However, some recommended commands (e.g., `find / -name "*.json" | grep token`) target the entire filesystem rather than the stated workspace and are broader than needed for onboarding.
Instruction Scope
SKILL.md and the docs instruct agents to read and write many local files (MEMORY.md, .learnings, memory/*), run filesystem searches, call platform tools (feishu_doc, cron list) and use network APIs (reddit, bocha). While most of this fits a how-to manual for agents, explicit guidance to search the whole filesystem for tokens or to run `find /` is a scope creep and raises risk of scanning sensitive system files. The docs do not instruct sending discovered secrets externally, but they do not restrict how found credentials are handled.
Install Mechanism
Instruction-only skill with no install spec and no code files — nothing is written to disk by an installer and no arbitrary downloads are specified. This is the lowest install risk.
Credentials
The registry metadata declares no required env vars or config paths, but the instructions reference accessing local workspace files and searching system files for tokens/configs. That mismatch (no declared config requirements but explicit advice to enumerate tokens/configs) is noteworthy: the skill implicitly expects access to local files even though it declares none.
Persistence & Privilege
Flags: always=false and disable-model-invocation=false (normal). The skill does not request permanent presence or modify other skills. No elevated privileges or persistence mechanisms are requested.
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 agent-training-manual - 安装完成后,直接呼叫该 Skill 的名称或使用
/agent-training-manual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
v1.0.0
悠悠大师姐的 AI 代理新员工培训手册 v1.0 — 30 天实战经验,8 章完整培训体系
元数据
常见问题
AI Agent Training Manual 是什么?
AI 代理新员工培训手册。包含心态原则、工作区规范、沟通规范、工具使用、定时任务、血泪教训、推荐技能清单、入职检查清单。适用于新 OpenClaw 代理快速上手,避免常见错误。当新代理加入团队、需要培训指导、或想了解 AI 代理最佳实践时使用。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 90 次。
如何安装 AI Agent Training Manual?
在 OpenClaw 或 Claude Code 对话框中运行命令「/install agent-training-manual」即可一键安装,无需额外配置。
AI Agent Training Manual 是免费的吗?
是的,AI Agent Training Manual 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。
AI Agent Training Manual 支持哪些平台?
AI Agent Training Manual 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 AI Agent Training Manual?
由 harrot90-code(@harrot90-code)开发并维护,当前版本 v1.0.0。
推荐 Skills