← 返回 Skills 市场
73
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install multi-agent-communication
功能描述
Based on two core tools, sessions_spawn and sessions_send, to help users build, manage, and optimize distributed Agent systems, enabling task decomposition,...
安全使用建议
This skill reads like an internal design/usage doc for a multi-agent orchestration API — it is coherent with that purpose, but several items need clarification before you install or enable it:
- Ask the author/platform: what platform services does this expect (Gateway RPC, SubagentRegistry, Sweeper) and what credentials/permissions are required? Right now the skill declares none, which is surprising given the described integrations.
- Watch defaults: the provided config shows agentToAgent.allow: ["*"]. If applied as-is, that allows unrestricted cross-agent communication. Require a whitelist or tighter defaults in production.
- Invisible child runs: mode="run" is described as completely invisible to the user. Confirm logging, auditing, and limits so background tasks cannot perform covert work on sensitive data.
- Attachment snapshotting: children get copies of attachments that persist independently. Ensure data retention, access controls, and deletion policies are explicit to avoid unintended data duplication/exfiltration.
- Require visibility controls: mandate thread binding or user-visible sessions for any task that touches sensitive data, and enable timeouts/limits (max depth, max children) appropriate for your environment.
- Test in isolation: run the patterns in a sandboxed environment first, and verify there are no unexpected network endpoints or external integrations being contacted.
If the maintainer can provide explicit details on required platform permissions, auditing/logging behavior, and recommended secure config (non-permissive defaults), re-evaluate; until then treat the skill as potentially risky for sensitive environments.
功能分析
Type: OpenClaw Skill
Name: multi-agent-communication
Version: 1.0.0
The skill bundle is a documentation-only package providing instructions and design patterns for multi-agent communication using the 'sessions_spawn' and 'sessions_send' tools. It includes configuration references (config.md) and architectural patterns (patterns.md) for parallel processing and agent collaboration. No executable code or malicious instructions were found; the content is entirely aligned with its stated purpose of managing distributed agent systems.
能力评估
Purpose & Capability
The name/description match the provided instructions: the skill documents two actions (sessions_spawn and sessions_send) and patterns for spawning and messaging Agents. However, the doc references binding to third‑party threads (Discord/Slack), a Gateway RPC, SubagentRegistry, and a Sweeper service — legitimate platform components, but the SKILL.md does not declare what platform privileges, credentials, or integrations are required for those features.
Instruction Scope
Instructions permit invisible 'run' mode child Agents (user-invisible execution) and describe copying attachments (snapshots) into child Agents' independent lifecycles. The doc also describes automatic push mechanisms (Sweeper polling every 1–8s) and thread binding. Those behaviors expand the agent's runtime footprint (background work, duplicated data, cross-agent access) and could be abused for covert processing or unintended data replication if not constrained or audited.
Install Mechanism
Instruction-only skill with no install spec and no code files — nothing is written to disk by the package itself, which minimizes installation risk.
Credentials
The skill declares no required environment variables or credentials, yet describes operations that in a real deployment typically require permissions (posting/reading from chat threads, invoking gateway RPCs, accessing registries, reading/writing shared workspaces). The references/config.md also shows a permissive default (agentToAgent.allow: ["*"]) which would allow all cross-agent communication by default — a risky default if applied in production.
Persistence & Privilege
always:false (good) and autonomous invocation is allowed (the platform default). The ability to spawn persistent 'session' Agents and to run invisible 'run' Agents increases runtime privilege and persistence within the platform; this is a legitimate capability but should be paired with auditing, visibility controls, and strict whitelists to avoid abuse.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install multi-agent-communication - 安装完成后，直接呼叫该 Skill 的名称或使用
/multi-agent-communication触发 - 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
v1.0.0
- Initial release of multi-agent-communication skill.
- Introduces two core tools: sessions_spawn (spawns child Agents) and sessions_send (communicates with existing Agents).
- Supports distributed task decomposition, parallel processing, and efficient inter-Agent coordination.
- Provides robust security controls: recursion depth, child process limits, and cross-Agent whitelisting.
- Enables multi-round agent-to-agent negotiation and supports both one-shot and persistent session modes.
元数据
常见问题
Multi-Agent Communication 是什么？
Based on two core tools, sessions_spawn and sessions_send, to help users build, manage, and optimize distributed Agent systems, enabling task decomposition,... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 73 次。
如何安装 Multi-Agent Communication？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install multi-agent-communication」即可一键安装，无需额外配置。
Multi-Agent Communication 是免费的吗？
是的，Multi-Agent Communication 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Multi-Agent Communication 支持哪些平台？
Multi-Agent Communication 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 Multi-Agent Communication？
由 OpenLark（@openlark）开发并维护，当前版本 v1.0.0。
推荐 Skills