hermes - 💡(How to fix) Fix [Feature]: Support dynamic role loading for sub-agents

Problem or Use Case / 问题或使用场景

Currently, sub-agents cannot dynamically load roles/personas to execute tasks. Each sub-agent must be configured with a fixed role at creation time, which limits flexibility in dynamic task assignment.

目前子代理无法动态加载角色/人物设定来执行任务。每个子代理必须在创建时配置固定的角色，限制了动态任务分配的灵活性。

Use case / 使用场景：

• Main agent breaks down a complex task into sub-tasks with different specialized roles / 主智能体将复杂任务分解为需要不同专业角色的子任务
• Dynamically assign coding expert, reviewer, documentation writer roles to different sub-agents / 动态为不同子代理分配代码专家、审查员、文档撰写者等角色
• Tasks requiring context switching between different personas / 需要在不同人物设定间切换的任务

Proposed Solution / 建议方案

Allow sub-agents to load roles dynamically at runtime:

允许子代理在运行时动态加载角色：

Option 1: Role Loading via Task Assignment / 方案1：任务分配时加载角色

Main agent assigns: "Use the 'security-expert' role to audit this code"
Sub-agent loads: roles/security-expert.md
Executes task with loaded persona

Option 2: Role Manifest / 方案2：角色清单

# Config or command
sub_agent:
  available_roles:
    - security-expert
    - code-reviewer
    - documentation-writer
    - domain-expert
  
  load_role: security-expert  # loaded before task execution

Option 3: Runtime Role Switching / 方案3：运行时角色切换

> /role security-expert
[Sub-agent persona switched to security-expert]

> /role code-reviewer  
[Sub-agent persona switched to code-reviewer]

Alternatives Considered / 备选方案

• Hard-coded role in sub-agent config: inflexible, requires recreation of agents / 子代理配置中硬编码角色：不灵活，需重新创建代理
• System prompt injection per message: fragile, easy to forget / 每条消息注入系统提示：脆弱，容易遗漏

Feature Type / 功能类型

• Configuration option
• CLI improvement
• Agent Core improvement

Scope / 范围

• Medium (few files, < 300 lines)