360 Secure SkillHub
360 安全技能中心

multi-agent-conductor

根据任务描述与可选禁止事项清单,通过 Python 规划脚本将大任务分解为子任务、分配 agent 角色,并按有门禁原则为每个 agent 划定权限范围与禁止项,输出编排流程、边界清单与结构化 JSON 结果。

zhaoxinghua09-cell
zhaoxinghua09-cell
热度 62.84
最近更新 2026-10-17 07:22:43
分类:协作依赖:代码执行 / 读写文件应用场景:任务管理 / 工作流

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add multi-agent-conductor
Metadata
name
multi-agent-conductor
display_name
多智能体编排(Multi-Agent Conductor)
display_name_en
Multi-Agent Conductor
description
当用户说『这个任务好大要拆给多个AI』『多智能体怎么分工』『agent之间怎么不串权限』『编排几个agent协作』,或要把一个大任务拆成多个 agent 并行/串行协作时使用。把任务分解为子任务→分配角色→划清每个 agent 的边界与禁止项(有门禁),输出编排方案+边界清单。可运行脚本(conductor_plan 规划器)。理论根基:LGD 三律之有门禁(任务边界+权限隔离)。触发词:多智能体、multi-agent、agent编排、任务分解、协作agent、agent权限、并行agent、orchestration。
version
1.1.1
agent_created
true
author
诺声(Logos)@SynomosAI
license
MIT
license_scope
代码(.py 文件)MIT;本 SKILL.md 与其中理论文本不在 MIT 覆盖范围内(见文末「License & attribution」)
allowed-tools
Read, Glob, Grep, Task
category
AI工程方法
platforms
["windows","macos","linux"]
platform_compatibility
["windows","macos","linux"]
read_when
["一个大任务要拆给多个 agent 协作","多 agent 出现权限串台/互相改对方产物","要设计 agent 团队的分工与边界","担心某个 agent 越权访问不该碰的资源"]
tags
["多智能体","multi-agent","编排","任务分解","协作","有门禁"]
slug
multi-agent-conductor
title
多智能体编排(Multi-Agent Conductor)
copyright
SynomosAI
description_zh
当用户说『这个任务好大要拆给多个AI』『多智能体怎么分工』『agent之间怎么不串权限』『编排几个agent协作』,或要把一个大任务拆成多个 agent 并行/串行协作时使用。把任务分解为子任务→分配角色→划清每个 agent 的边界与禁止项(有门禁),输出编排方案+边界清单。可运行脚本(conductor_plan 规划器)。理论根基:LGD 三律之有门禁(任务边界+权限隔离)。触发词:多智能体、multi-agent、agent编排、任务分解、协作agent、agent权限、并行agent、orchestration。
description_en
Decompose large tasks into multi-agent orchestration: subtask split, role assignment, per-agent permission boundaries, and parallel or serial coordination.
classification
{"skillhub":["security-compliance"],"iso_25010":["Security"]}
risk_tier
medium
risk_rationale
门禁批量补丁默认 medium(内部约定:本 skill 设计为中等风险):调用 API/读写文件/本地执行,最小权限,密钥走 Secret;若实际涉对外代执行(push/发布/发送/上架)须人工复评,改 high 并走负责人批准。
ai_protocol
true
spec
usk/1.0
interface
{"type":"cli","entry_point":"scripts/conductor_plan.py","runtime":"python3","call_pattern":"args"}
input_schema
{"task":{"type":"string","description":"task","required":true},"agents":{"type":"int","description":"agents","required":false},"json":{"type":"bool","description":"json","required":false}}
output_schema
{"stdout":"结构化 JSON IR(含 result / rc / error_code / warnings)","exit_code":"rc=0 正常 / rc=1 告警或从严推定 / rc=2 输入不足或非法"}
capabilities
["multi-agent","orchestration"]
permissions
{"network":false,"filesystem":false,"subprocess":false}
requirements
{"python_packages":[],"min_python":"3.8"}
changelog
1.1.1: Bilingual rewrite (full EN+ZH, English first); 1.0.0: initial upload (USK v3)

凡自治之物 · LGD 治理学 · 创造术家族 | 评测协议:skill-quality-gate

English

When to use

  • When the user says a task is too large for a single agent and must be split among multiple AIs ("this task is too big, split it across several AIs").
  • When permission isolation and responsibility boundaries between multiple agents matter, to prevent overstepping or collusion ("how do agents not leak permissions into each other").
  • When you need to define the orchestration mode (parallel / serial) and the subtask allocation, in a way that complies with the LGD gated principle.

Usage

  • Tool: conductor_plan (backed by conductor_plan.py).
  • Invocation: python conductor_plan.py --mode orchestration --input task_description.
  • Parameters:
    • --mode: parallel or serial.
    • --input: task-description text or structured JSON containing the core goal and constraints.
    • --permission: (optional) a prohibited-items list to strengthen the gate rules.

Workflow

  1. Trigger detection: spot keywords such as "multi-agent" or "task decomposition" and decide whether to intervene.
  2. Requirement analysis: extract the core task and assess a single agent's capability boundary and potential risks.
  3. Script execution: run conductor_plan.py to decompose the logic and produce the subtask list and role assignment.
  4. Boundary drawing: per the LGD gated principle, define each agent's permission scope and prohibited items.
  5. Plan generation: output the final report containing the orchestration flow and the gate checklist.
  6. Feedback: return the result to the user and ask whether the execution strategy needs adjustment.

Evidence & safety statement

  • Evidence: TBD (no evaluation / badge link yet; update here once produced)
  • Version: 1.1.1
  • Safety: no independent security scan has been performed (TBD); the conclusion is pending and no endorsement is given until it is completed.

中文

何时用

  • 用户提出任务庞大、单一 Agent 无法完成,需拆分给多个 AI 协同时(如“这个任务好大要拆给多个 AI")。
  • 涉及多智能体间的权限隔离、职责边界划分,防止越权协作或串通风险时(如"agent 之间怎么不串权限”)。
  • 需要明确 Agent 的编排模式(并行/串行)及子任务分配方案,且需符合 LGD 三律之有门禁原则时。

用法

  • 工具: conductor_plan (基于文件 conductor_plan.py)。
  • 调用命令: python conductor_plan.py --mode orchestration --input task_description.
  • 参数说明:
    • --mode: 可选值,指定 parallel(并行)或 serial(串行)。
    • --input: 任务描述文本或结构化 JSON,需包含核心目标与约束。
    • --permission: (可选)定义禁止事项清单,用于强化门禁规则。

流程

  1. 触发检测: 识别用户输入中的“多智能体”、“任务分解”等关键词,判断是否需要介入。
  2. 分析需求: 提取核心大任务,评估单一 Agent 的能力边界与潜在风险。
  3. 执行脚本: 启动 conductor_plan.py 进行逻辑拆解,生成子任务列表及角色分配。
  4. 划定边界: 依据 LGD 三律之有门禁原则,明确每个 Agent 的权限范围与禁止项。
  5. 生成方案: 输出包含编排流程与门禁清单的最终报告。
  6. 反馈确认: 将结果返回给用户并询问是否需要调整执行策略。

证据与安全声明

  • 证据:TBD(暂无评测/徽章链接;产生后在此更新实际链接)
  • 版本:1.1.1
  • 安全:未做独立安全扫描(TBD),结论待补;补全前不作推荐背书。