← 返回 Skills 市场
krislu1221

Agent-Selector

作者 Krislu · GitHub ↗ · v1.0.3 · MIT-0
cross-platform ✓ 安全检测通过
193
总下载
0
收藏
0
当前安装
3
版本数
在 OpenClaw 中安装
/install agent-selector
功能描述
自动根据任务描述在146+专业Agent之间安全切换，支持自动适配与手动指定，确保最适合的人格身份工作。
使用说明 (SKILL.md)

Agent Selector Skill v2.0 - 人格身份切换

📋 技能说明

全局技能，基于 Agency-Agent 人格库的优化，在不同专家人格之间自动切换，支持 146+ 个专业 Agent。根据对话自动适配，判定任务结束自动切换回初始人格，无须手动频繁切换。避免人格的混乱，同时始终保持最适合的人格工作。

该功能也集成在 Auto-coding 和 RoundTable skill 中。

✨ v2.0 安全增强

  • ✅ 路径白名单验证 - 防止目录遍历攻击
  • ✅ 文件大小限制 - 100KB 限制，防内存耗尽
  • ✅ 输入长度限制 - 10000 字符限制，防 DoS
  • ✅ 符号链接验证 - 防止访问外部文件
  • ✅ 文件编码验证 - 防止编码注入
  • ✅ 错误信息脱敏 - 防止信息泄露
  • ✅ 只读操作 - 不写入外部目录，无安装脚本
  • ✅ 范围明确 - 仅限本地 Agent 选择和 Prompt 加载

🔒 安全范围声明

本技能的功能范围：

  • ✅ 读取 bundled 的 Agent Prompt 文件
  • ✅ 根据关键词自动选择 Agent
  • ✅ 缓存已加载的 Prompt
  • ✅ 任务完成后恢复默认身份

本技能不执行的操作：

  • ❌ 不写入外部目录
  • ❌ 不创建符号链接
  • ❌ 不安装到其他工具
  • ❌ 不请求外部 API
  • ❌ 不持久化存储

🚀 快速开始

方式 1：自动选择 Agent

from agent_selector_skill import auto_select_agent

# 根据任务自动匹配合适的 Agent
agent = auto_select_agent("设计 React 前端架构")
# 返回：engineering/engineering-frontend-developer

方式 2：手动切换人格

from agent_selector_skill import AgentSelector

selector = AgentSelector()

# 加载特定 Agent 的 Prompt
prompt = selector.load_agent_prompt("engineering/engineering-frontend-developer")

# 现在你以"前端工程师"人格思考

方式 3：RoundTable 专用

from agent_selector_skill import select_roundtable_agents

# 为 RoundTable 讨论选择合适的 Agent
agents = select_roundtable_agents("智能客服系统技术方案")
# 返回：['engineering/engineering-frontend-developer', ...]

🎭 可用人格分类

Engineering（工程类）

  • engineering/engineering-frontend-developer - 前端开发工程师
  • engineering/engineering-backend-developer - 后端开发工程师
  • engineering/engineering-fullstack-developer - 全栈开发工程师
  • engineering/engineering-software-architect - 软件架构师
  • engineering/engineering-devops-automator - DevOps 工程师
  • engineering/engineering-security-engineer - 安全工程师

Testing（测试类）

  • testing/testing-qa-engineer - QA 工程师
  • testing/testing-accessibility-auditor - 可访问性审计师

Design（设计类）

  • design/design-ux-designer - UX 设计师
  • design/design-ui-designer - UI 设计师
  • design/design-interaction-designer - 交互设计师

Product（产品类）

  • product/product-manager - 产品经理

Specialized（专业类）

  • specialized/specialized-ai-ml-engineer - AI/ML 工程师

完整列表: 146+ 个 Agent，见 agency-agents/ 目录

📊 使用场景

场景 1：日常对话切换

用户：帮我设计一个前端架构

你：（自动切换到前端工程师人格）

✅ 已切换到前端工程师人格

## 技术方案

基于 React 18 + TypeScript 的前端架构...

场景 2：RoundTable 多 Agent 讨论

from roundtable_skill import RoundTableEngine
from agent_selector_skill import select_roundtable_agents

# 自动选择合适的 Agent
agents = select_roundtable_agents("智能客服系统技术方案")

# 创建 RoundTable 引擎
engine = RoundTableEngine(
    topic="智能客服系统技术方案",
    agents=agents  # 使用自动选择的 Agent
)

# 执行讨论
await engine.run(user_channel)

场景 3：Auto-Coding 自主编码

from auto_coding import AutoCodingAgent
from agent_selector_skill import auto_select_agent

# 根据任务自动选择 Agent
agent_id = auto_select_agent("开发 Python REST API")

# 创建编码 Agent
agent = AutoCodingAgent(agent_id=agent_id)

# 执行任务
result = await agent.execute("开发用户管理 API")

🔒 安全特性

路径安全

# ✅ 允许的路径
AgentSelector()  # 使用内置 agency-agents
AgentSelector("agency-agents")  # 白名单目录

# ❌ 拒绝的路径
AgentSelector("/etc/passwd")  # 路径遍历攻击
AgentSelector("../secret")  # 相对路径攻击

文件大小限制

# 自动跳过超大文件（>100KB）
# 防止内存耗尽攻击

输入长度限制

# 任务描述限制 10000 字符
# 防止 DoS 攻击

符号链接验证

# 只允许内部符号链接
# 拒绝指向外部的符号链接

📁 文件结构

agent-selector-skill/
├── SKILL.md                  # 本文档
├── __init__.py               # 模块导出
├── agent_selector.py         # 核心选择器（安全增强版）
├── clawhub.json              # ClawHub 配置
└── README.md                 # 详细说明

🔧 配置说明

Agent 来源

# 方式 1：使用内置的 agency-agents（推荐）
selector = AgentSelector()

# 方式 2：使用外部 Agent（必须在白名单内）
selector = AgentSelector("agency-agents")

# ❌ 错误：路径不在白名单内
selector = AgentSelector("/path/to/external")  # 会抛出 ValueError

自动选择规则

# 关键词匹配
"react" → engineering/engineering-frontend-developer
"python" → engineering/engineering-backend-developer
"测试" → testing/testing-qa-engineer
"ux" → design/design-ux-designer
"ai" → specialized/specialized-ai-ml-engineer

🧪 测试

# 运行内置测试
cd \x3CYOUR_OPENCLAW_WORKSPACE>/skills/agent-selector-skill
python3 agent_selector.py

预期输出:

============================================================
Agent Selector - 安全增强版测试
============================================================

📊 可用 Agent 数量：146

============================================================
任务匹配测试：
============================================================

任务：设计一个 React 前端架构
匹配 Agent: engineering/engineering-frontend-developer
------------------------------------------------------------
...

✅ 所有测试完成！

📚 集成示例

RoundTable 集成

已在 roundtable-skill/roundtable_engine.py 中集成：

from agent_selector import AgentSelector, select_roundtable_agents

class RoundTableEngine:
    def __init__(self, topic: str, ...):
        self.agent_selector = AgentSelector()
        self.agents = select_roundtable_agents(topic)

Auto-Coding 集成

已在 auto-coding/agent_controller.py 中集成：

from agent_selector import AgentSelector

class AgentController:
    def __init__(self):
        self.agent_selector = AgentSelector()
    
    def select_agent_for_task(self, task: str):
        return self.agent_selector.select_agent(task)

⚠️ 注意事项

1. 路径限制

  • 只能访问 skills 目录内的 Agent
  • 目录名必须在白名单中（agency-agents, agency-agents-zh, agency）

2. 文件大小

  • 单个 Agent 文件不能超过 100KB
  • 超大文件会被自动跳过

3. 错误处理

try:
    selector = AgentSelector("/invalid/path")
except ValueError as e:
    print(f"路径验证失败：{e}")

🔗 相关链接

📝 更新日志

v2.0.0 (2026-03-20) - 安全增强版

  • ✅ 添加路径白名单验证
  • ✅ 添加文件大小限制（100KB）
  • ✅ 添加输入长度限制（10000 字符）
  • ✅ 添加符号链接验证
  • ✅ 添加文件编码验证
  • ✅ 错误信息脱敏
  • ✅ 预编译正则表达式（防 DoS）

v1.0.0 (2026-03-19) - 初始版本

  • 基础 Agent 选择功能
  • 关键词匹配
  • RoundTable 集成

📄 许可证

MIT License - 虾软 Claw soft

Agent Selector - 让你成为任何需要的专家

安全使用建议
This skill appears internally consistent with its purpose: it reads bundled prompt files and selects agents based on keywords, with path/size/encoding checks. Before installing, verify you place the skill under your workspace's skills/ directory (so its white‑list protections work), quickly skim agent_selector.py for any unexpected network or subprocess calls (the truncated file you saw showed none), and run it in a sandboxed workspace first if you want extra assurance. If you need higher assurance, inspect the remainder of agent_selector.py (the full implementation) for any network/socket usage or calls to os.environ not described in the docs.
功能分析
Type: OpenClaw Skill Name: agent-selector Version: 1.0.3 The Agent Selector skill is a persona management utility designed to switch agent identities based on conversation context. The core logic in `agent_selector.py` is well-hardened, featuring explicit protections against path traversal (via `_validate_agent_source` and `skills_root` resolution), symlink attacks, and DoS (via strict file size and input length limits). The extensive library of agent personas in the `agency-agents/` directory consists of professional system prompts aligned with their stated expert roles. No evidence of malicious execution, data exfiltration, or harmful prompt injection was found.
能力评估
✓ Purpose & Capability
Name/description (Agent Selector for switching between 146+ local Agents) match the included assets and code: the package contains a large bundled 'agency-agents' prompt library plus a selector implementation. Nothing in the metadata or code requires unrelated cloud credentials or external services.
✓ Instruction Scope
SKILL.md and the visible code only describe reading bundled prompt files, keyword-based selection, caching, and returning agent IDs/prompts. The instructions and examples reference only local, whitelisted directories and include path/size/encoding/symlink checks. There are no instructions that ask the skill to read arbitrary system files or to exfiltrate data.
✓ Install Mechanism
No install spec or remote downloads are present; this is effectively an instruction/package bundle that the user copies into their skills folder. All files are bundled in the package (no external URL downloads or extract steps), which is proportionate for a prompt library + selector.
✓ Credentials
The skill requires no environment variables/credentials and the SKILL.md declares read-only filesystem access. There are no unexpected secret-like env var names or config paths requested; the claimed operations (reading prompt files) align with the requested access.
✓ Persistence & Privilege
always is false (no forced global inclusion) and the skill states it performs read-only operations and does not persist external state. The package contains only its own files and documentation and does not appear to modify other skills or system-wide config.
如何使用
  1. 确保已安装 OpenClaw（本地或 Docker 部署）
  2. 在对话框中输入安装命令：/install agent-selector
  3. 安装完成后，直接呼叫该 Skill 的名称或使用 /agent-selector 触发
  4. 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
v1.0.3
- Added LICENSE file to the agency-agents subdirectory. - No other functional or code changes; this update clarifies licensing for bundled agents.
v1.0.2
- Skill 名称由 "agent-switcher" 统一调整为 "agent-selector"。 - 所有导入/使用示例及相关文档说明同步更名为 agent_selector_skill。 - 配置、测试、集成等说明中的 skill 目录、类与方法名统一按新名称更新。 - ClawHub 链接等外链也已更新为新 skill 名称。 - 其余功能、API、目录结构未变。
v1.0.1
Agent Switcher Skill v1.0.1 - Adds comprehensive safety enhancements: path whitelist validation, file size (100KB) and input length (10,000 chars) limits, symlink and encoding checks, and sensitive error masking. - Prevents directory traversal, memory exhaustion, external file access, and DoS attacks. - Documentation updated with new usage examples, scenario guides, and integration references. - Agent selection logic and categorization remain, now with secure handling throughout. - Backward compatible with earlier agent selection and RoundTable integration features.
元数据
Slug agent-selector
版本 1.0.3
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 3
常见问题

Agent-Selector 是什么？

自动根据任务描述在146+专业Agent之间安全切换，支持自动适配与手动指定，确保最适合的人格身份工作。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 193 次。

如何安装 Agent-Selector？

在 OpenClaw 或 Claude Code 对话框中运行命令「/install agent-selector」即可一键安装，无需额外配置。

Agent-Selector 是免费的吗？

是的，Agent-Selector 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。

Agent-Selector 支持哪些平台？

Agent-Selector 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 Agent-Selector？

由 Krislu（@krislu1221）开发并维护，当前版本 v1.0.3。

推荐 Skills
self-improving agent
Captures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Clau...
⬇ 463283 · by pskoett
Skill Vetter
Security-first skill vetting for AI agents. Use before installing any skill from ClawdHub, GitHub, or other sources. Checks for red flags, permission scope, and suspicious patterns.
⬇ 259410 · by spclaudehome
Polymarket
Query Polymarket prediction markets. Check odds, find trending markets, search events, track price movements.
⬇ 232503 · by joelchance
Self-Improving + Proactive Agent
Self-reflection + Self-criticism + Self-learning + Self-organizing memory. Agent evaluates its own work, catches mistakes, and improves permanently. Use when...
⬇ 200423 · by ivangdavila
Github
Interact with GitHub using the `gh` CLI. Use `gh issue`, `gh pr`, `gh run`, and `gh api` for issues, PRs, CI runs, and advanced queries.
⬇ 191113 · by steipete
ontology
Typed knowledge graph for structured agent memory and composable skills. Use when creating/querying entities (Person, Project, Task, Event, Document), linkin...
⬇ 190156 · by oswalpalash

💬 留言讨论