Agency Orchestrator

多 Agent 协作调度系统 - 自动分析任务、选择最佳 Agent、协调多 Agent 协作完成复杂任务

Security Scan

The skill's code mostly matches its orchestration purpose, but embedded files contain a hardcoded import of user-specific paths and an integration script that persistently modifies your shell/profile — these are incoherent or risky and merit review before installing.

Purpose & Capability

The name/description (multi-agent orchestrator) aligns with the provided Python modules: scanning ~/.openclaw/agency-agents-zh for agent .md files, analyzing tasks by keywords, selecting agents, and logging interactions. However, some included files (collaborative_mode.py) reference external modules and absolute user-specific paths that are not explained by the skill's description.

Instruction Scope

SKILL.md and README consistently reference config and log files under ~/.openclaw/agency-agents-zh which matches the code. But the repository contains a module (collaborative_mode.py) that inserts hardcoded absolute paths (/Users/ghq/...) into sys.path and imports CollaborativeAgentSystem from an external module — this causes the skill to reach into arbitrary user directories and execute code outside the skill bundle if that file is run/imported. The provided integrate_with_clawx.sh script also edits your ~/.bash_profile and openclaw.json to enable the skill, which is a persistent, privileged operation not described in a way that highlights risk.

Install Mechanism

There is no automatic install spec; the skill is instruction/code-only. No remote downloads or archive extraction are performed by the package itself. The only installation helper is an included shell script (integrate_with_clawx.sh) that edits local config files and shell profile — it must be run manually.

Credentials

The skill declares no required environment variables or credentials and the code does not attempt to read secrets. The integrate script adds an AGENCY_DIR environment variable to ~/.bash_profile and adds that directory to PATH, which is reasonable for convenience but unnecessary for core functionality and increases persistence surface.

Persistence & Privilege

always is false and the skill does not demand autonomous invocation privileges. The integrate_with_clawx.sh script will (if run) modify your ~/.bash_profile and enable the skill in ~/.openclaw/openclaw.json, creating persistent changes and adding the agency directory to PATH; this is a significant local change and should be reviewed before execution.

What to consider before installing

This package generally implements an agent-orchestration feature that matches its description, but there are two things you should check before installing or running anything: (1) open collaborative_mode.py — it inserts hardcoded absolute paths (/Users/ghq/...) and imports an external collaborative_agents module. If those paths exist on your machine they could cause arbitrary code execution when that module is imported; if they don't exist the import will fail. Consider removing or sandboxing this file or at least removing the hardcoded sys.path inserts. (2) inspect integrate_with_clawx.sh before running: it will edit your ~/.bash_profile (adding AGENCY_DIR and PATH entries) and write to ~/.openclaw/openclaw.json to enable the skill. Back up those files first, and run the script only after reviewing/modifying it to match your environment. Also manually inspect any agent .md files placed under ~/.openclaw/agency-agents-zh for unexpected content. If you want higher assurance, run the skill in a disposable account/container or ask the author for clarification about collaborative_mode.py and why it needs to import from /Users/ghq.

Agency Orchestrator - 多 Agent 协作调度系统

功能

• 🧠 智能任务分析 - 自动分析任务需求，识别所需技能
• 🤖 Agent 自动选择 - 从 166+ 个 Agent 中选择最佳组合
• 🤝 协作协调 - 管理多 Agent 通信和任务交接
• 📚 持续学习 - 从每次交互中学习，优化 Agent 表现

使用方式

基础使用

from agency_orchestrator import AgentOrchestrator

orchestrator = AgentOrchestrator()
result = orchestrator.coordinate("帮我设计一个产品官网")

在 ClawX 中使用

qwen -p "使用 agency-orchestrator 完成以下任务：设计电商网站"

可用 Agent 分类

• design (8 个) - UI/UX 设计
• engineering (24 个) - 软件开发
• marketing (31 个) - 营销策略
• sales (8 个) - 销售支持
• product (5 个) - 产品管理
• testing (8 个) - 质量保证
• ... 25+ 分类

配置

配置文件：~/.openclaw/agency-agents-zh/config/settings.json

日志

• 调度日志：~/.openclaw/agency-agents-zh/logs/orchestrator_log.json
• 学习日志：~/.openclaw/agency-agents-zh/logs/learning_log.json
• 协作日志：~/.openclaw/agency-agents-zh/logs/collaboration_log.json