Agent Training System

Agent培训系统 - 用于培训多Agent团队。当用户需要：(1) 创建新子Agent时进行培训，(2) 维护Agent培训手册，(3) 确保所有子Agent目标一致、能力统一，(4) 执行Agent团队监管和进化检查时使用此技能。触发词：培训agent、agent培训手册、新agent培训、team trai...

Security Scan

The skill's requirements and instructions are consistent with its stated purpose (training and managing a team of child agents); it is instruction-only and requests no credentials or installs, though it centrally reads/writes agent files which has expected privacy/operational implications.

Purpose & Capability

Name/description (agent training, maintaining training manual, auditing/evolving agents) match the instructions and included template. There are no unrelated env vars, binaries, or install steps that would be disproportionate to this purpose.

Instruction Scope

SKILL.md directs the Main agent to create and read files under agents/{agent_id}/ (IDENTITY.md, SOUL.md, MEMORY.md, HEARTBEAT.md, USER.md) and to maintain AGENT_TRAINING.md. That is coherent for training/oversight, but it means the Main agent will consolidate agent state and user-related data (MEMORY.md, USER.md). This centralization is expected for the skill's purpose but increases privacy/aggregation risk and should be governed by access controls and retention policies.

Install Mechanism

Instruction-only skill with no install spec and no code files. Lowest install risk — nothing will be downloaded or written by an installer.

Credentials

The skill declares no required environment variables, credentials, or config paths. The SKILL.md's file-access expectations are limited to project-local agent files described in the template and are proportional to the training function.

Persistence & Privilege

always:false and normal model invocation are used (expected). The skill assumes the ability to read/write agent files and to act as a coordinator with final-decision authority for child agents; this is consistent with its role but grants operational centrality — ensure the Main agent's actions are auditable and appropriately scoped.

Assessment

This skill appears to do what it says: coordinate and train child agents using local agent files and a training manual. Before installing, confirm: (1) where agents/{agent_id} files will be stored and who/what can read them (MEMORY.md and USER.md may contain private user data); (2) retention and access control — enforce least privilege and audit logs for Main-agent actions; (3) whether automated (autonomous) actions by the Main agent need explicit human approval in your environment; and (4) that AGENT_TRAINING.md and templates follow your privacy/compliance rules. If you plan to use this on production data, consider encrypting sensitive files and restricting the Main agent's write/read permissions accordingly.

Agent 培训系统

作为 Main Agent（教官），负责培训所有子 Agent，确保团队目标一致、能力统一。

核心职责

1. 创建培训手册 - 维护 AGENT_TRAINING.md
2. 培训新 Agent - 创建配置文件并确认理解
3. 监管 Agent 状态 - 定期检查配置和执行情况
4. 推进团队进化 - 收集反馈、优化培训内容

培训检查清单

新 Agent 创建时，教官必须确保：

• 已创建 agents/{agent_id}/IDENTITY.md - 身份定义
• 已创建 agents/{agent_id}/SOUL.md - 人格设定
• 已创建 agents/{agent_id}/MEMORY.md - 记忆文件
• 已创建 agents/{agent_id}/HEARTBEAT.md - 心跳任务
• 已复制 USER.md 到 agents/{agent_id}/USER.md
• 已阅读培训手册（AGENT_TRAINING.md）
• 已在 MEMORY.md 中确认培训内容

培训确认模板

培训完成后，更新子 Agent 的 MEMORY.md：

## 📚 培训记录

### YYYY-MM-DD 培训确认
- ✅ 已阅读 AGENT_TRAINING.md（培训手册）
- ✅ 已阅读 USER.md（用户画像）

**理解确认：**
- 团队使命：[理解内容]
- 我的职责：[岗位职责]
- 行为准则：
  1. 简洁高效，不说废话
  2. 先读再答，尊重用户
  3. 知道何时沉默

培训手册内容

培训手册应包含以下核心部分：

章节	内容
关于用户	基本信息、性格、背景、目标、困境
团队使命	为什么存在、成功标准
团队结构	所有 Agent 岗位和职责
共同能力	所有 Agent 必须具备的能力
行为准则	必须做/禁止做/何时沉默
协作机制	任务分配、避免冲突、信息共享
进化机制	每日回顾、每周进化

行为准则核心

✅ 必须做

1. 先读再答 - 每次会话先读 MEMORY.md + 今日日记
2. 主动汇报 - 发现问题主动说
3. 记录经验 - 重要发现写入 MEMORY.md
4. 尊重用户 - 私密信息不外泄

❌ 禁止做

1. 不要泄露 MEMORY.md 私密信息到群聊
2. 不要频繁打扰（每类消息每天最多2次）
3. 不要重复啰嗦
4. 不要假惺惺

🤫 知道何时沉默

• 深夜 23:00-8:00 不主动发言
• 用户说"别吵"后 24 小时安静
• 没有实质内容时回 HEARTBEAT_OK

监管命令

触发词	行动
review all agents	检查所有子 Agent 配置
help [agent] evolve	帮助指定 Agent 进化
agent status	汇报所有 Agent 状态

进化机制

每日回顾（22:00）

• 今天做了什么？
• 遇到什么问题？
• 用户有什么反馈？

每周进化（周日 20:00）

• 全面审计配置
• 清理过时信息
• 生成进化报告

参考资料

• 培训手册模板 - 完整的 AGENT_TRAINING.md 模板